STUDIES ON DESTRUCTIVE AND NON-DESTRUCTIVE QUALITY EVALUATIONS F MANGO FRUITS
ABSTRACT ES SUBM RRTAWARD OFTiE.DEGREEOF 9
Doc$
POST N NEE I D= TE H L Y
]k' I II'II U 1 P t L
ler.t le u perviSion•of-. Dr. Mohammad All Khan -- Dr. Abfl Jit Kar Departrncnt of Post Harvest DIVisLon of Food Seance & Engineering & Techrycio9Y Post Harvest Technology Aligarh Muslim University, Aligarh IA RI, New Delhi -110 012
DEPARTMENT OF POST HARVEST EN INEERU G AND TECHNOLOGY FACULTY OF AGRICULTURAL SCIENCES ALIGARH MUSLIM UNIVERSITY ALIGARH (INDIA} April 2014. 4¼1CV 2014 Abstract
Mango (Mangi /ra iridica L) is the most popular climacteric fruit and considcred as the king of fruits in India. India shares about 56% in the world production and ranks first in area and production of mango with 2,31 million ha and 12.75 million tonnes, respect~vei . In spite of being the world's largest producer of mango fruits, surprisingly, India accounts for less than one percent of the global trade because of ihe absence of rapid, -i ori-destructive and precision sorting methods for quality and safety assurance. In addition, only about 2 per cent of the fruits and vegetables in Irulia are processed, which is very low cornpared- -to -other countries. Lack of processing and in adequate storage facilities for fruits and vegetables- results in huge wastages (25-40%). Therefore, it is very necessary to improve qualit).r evihiaIioff techniques of food products to satisfy the increased awareness, sophistication, greater expectation of consumers and also to avoid huge losses. Objective methods (destructive methods and nondestructive) among the several techniques have been applied in this study,
Four Varieties Of mango (cv. CJrnrsa, Dnsh rnri, Langro and Malda) fruit from four orchards were selected for destructive quality evaluation study. Fruits were harvested at pre-, optimum and over maturity (when the majority of the mangoes were ripening on the tree) stages of ripening. All cof ected fruits were brought to laboratory in a temperature controlled system (1S°C) and stored immediately in a cold room maintained at 20°C, 95% RH, till further experimentation. Flotation method was used to confirm the maturity and for subsequent grading based on maturity. All the physical and hiochernical analyses of the fresh samples were conducted at ambient conditions (28 ± 0.1°C and 70 ± 1% RH) within 24 h of harvesting. The samples were then stored at 20±2°C and 95±1° relative humidity (RH) and analyses were repeated at the interval of 3 days up to 15 days of storage. '
On-the other had, for the non-destructive quality analysis of mango,. 60 ripe fruits of two cultivars (i.e. C)ausa and Dashehar i) were s elected in the spring season of 2011 on the basis of peel color which is more suitable for the image processing. Computer vision system based on four cameras (monochrome, color, IJY and NIR) was used in the present work one- by-one to check the -potential of image processing technique in sorting/grading of mango fruit Using the said cameras and two different lighting (ultra-violet and visible) regimes, six images
1 of each fruit were captured. The processing of acquired images was started with the transfer of the image to the specific buffers of the RAM memory of the PC and sub-sampled or trimmed by nearest neighbour- interpolation method to reduce the computational- time and expense of the whole process. Some image. pre-processing operations were also employed on the input image before applying the proposed physical characterization process. TMAQ Vision which is a libiasy of Lab IEW software was used in image processing, analysis and development of algorithms for morphological feature extraction as well as for defect detection of mango fruits.
The non-destructive method for quality evaluation of mango fruit based on computer vision technique generally follows certain definite steps such as image acquisition, image processing and segnleiltation. object measurement (feature extraction), and classification (decision-making). The first step (image acquisition) is considered as an important step for all above said camera based computer vision systems as this helps in processing of images and the separation of the object from the background (i.e, second step) and in the feature extraction (i.e. third step). The third step is generally considered as the, most significant step as it determines the overall performance of the image processing system. This method can be termed as successful if the output features c eseribe the processed data in a unique runner. Finally, the last step is the classification step in which analyzes the extracted features and the acquired knowledge is stored for fuuther classification of the data.
As evidamt from the experimental results, this type of non-destructive technique is a very useful technique for ana1ying the quality parameters which in turn help in sorting/grading of the fruit and assist in achieving the specified objectives prior to sending too mango fruit in the global market, Currently, subjective (organoleptic) or destructive or manual methods such as visual inspection, chemical testing, sensory evaluation or other laboratory based quality testing approaches, have been reported as the tools for quality monitoring of the fruits before exporting to international markets, especially in Asian countries such as India. These conventional techniques based on standard methodologies, are expensive, laboirous, time- consuming and prone to errors which result in huge losses not only in terms of economy and the quality of fruits but also in terms of reputation of fruit producers. For instance, many times export consignments from India are rejected even at the destination part due to detection
2 of soft tissue and some bacteria inside the mangoes or due to over and under ripeness (sweetr est).
Therefore, to overcome the inaccuracy and inefficiency, the image processing techniques based an computer vision system can be used as an efficient and real time alternative, tool/technique to support the conventional techniques for mango fruit sorting/grading. The principal aspects of this study are. (1) Studying the variability of mango quality at the harvest stage, during past harvest storage and due to cultivar difference_ (2) Studying the potential of four different cameras (monochrome, color, ultra-violet =d near-infra-red) in computer vision system for quality evaluation of mango. (3) Development of an algorithm for the extraction of various morphological features of mango fruits using each camera setup for Chausa and Dashehari cultivars. (4) Development of algorithms for the detection of the defects on the surface of mango fruits using monochrome and color (RIB) camera computer vision system. (5) Development of an algorithm for bruise detection just below the peel of mango fruits using near-infra-red camera based computer vision system_ (6) Investigation of various types of defects responsible for the Ioss of quality of mango and its market value using UV camera based computer vision system.
For high commercial value in national and international fruit markets and for likeness by people, the morphological features of mango fruits such as size, shape and surface defects are of great significance. The various factors such as harvest stage, cultivar, storage period and their combination (all two way interaction) have significant effect on physico-biochemical parameters (length, width, thickness, arithmetic mean diameter, geometric mean diameter, sphericity, aspect ratio, specific gravity, total soluble solids, titratable acidity, total carotenoids and color values) of mango fruit of four varieties (ev. Chausa, Dashe ari, I. ngra an id uk ia). However, the factor harvest stage as such and its interaction with the oultivar and with storage period, and the interaction between storage period and rultivar did not affect the length, aspect ratio and the sphericity of the mango fruit. The results computed with image processing (IF) based on different cameras and actual (Digital Vernier Caliper) method were compared and it was found that the coefficient of determination (1) for fruit length, width and thickness of Goth cultivars was between 0.925
3 and 0.948, 0.781 and 0.922, and 0.80'4 and 0.909, respectively. The proposed IP method yielded accuracies above 32%, 78 % and 80% in estimating mango fruit length, width and thickness, respectively fur all cameras used in the computer -vision system.
The second comparison was provided by means of the paired t-test results. The difference in mean values of dimension (length, width and thickness) measured by both methods and the corresponding standard deviation (SD) have been reported in this research. "Fie paired t-test results confirmed the dimension computed with IF method and [WC are statistically significant at the 5% level (P > 0.05). Further, to reveal a relationship between the differences and the averages, to look for any systematic bias and to identify -possible outliers, the Bland-Altman plot or the difference plot was also used and it was found that the mango fruit diameters have no effect on the accuracy of the estimated dimensions. This plot indicated that the diameters differences between the computed and actual diameters were normally distributed. The outliers indicate the 95°%4 limit of agreement and the centre line shows the average differences.
Mathematical models were developed for predicting mass and volume of the fruit using area, perimeter, Max Feret's and Waddel Disk diameters as the independent variables. The geometrical characteristics like major diameter (length), intermediate diameter (width) and minor diameters (thickness), and various derivative parameters (such as aspect ratio, sphericity, arithmetic and geometric diameter) in different combinations were employed to develop models to predict the mass and volume of fruits. Higher values of the eoef'f cient of correlation Were obtained for the models based on geometric attributes computed by the image processing method based monochrome computer vision system followed by UV and NIR and then color based computer vision system. The. models based on aspect ratio and sphericity (for all cameras and lighting regimes setup) correlated inadequately with the mass and volume of fruits,
The results obtained in the case of monochrome camera setup were found to be superior to those obtained from color camera, except in some cases that involved fruit with-minor defects. Similarly, the overall performance of the algorithm in the case of monochrome camera setup for defect detection was better than that for color camera scrap. The NIk based system was used to identify the black spots just below the fruit's peel and an algorithm is
4 proposed for this purpose in this research. On the other hand, several types of bruises (anthracuos•e, latex stain, mechanical damage, water core and black spot underneath of the peel, and recognition of shriveled fruits) on the fruit's surface and inside•the fruits were found visible in UV light when ultra-violet camera based computer vision system was used. I-0 addition, the results related to morphol>gicii features from LJV camera haisod computer vision system were consistent and exhibited a high coefficient of correlation.
Consequently at the end, the principal conclusion from this thesis report is that the monochrome camera based computer vision technique for the evaluation of mango fruit quality can be more successful if implemented to assist the traditional methods for quality monitoring before exporting to the.global market. However, all four cameras based computer vision systems were found to have great potential in detecting the defects of the mango Emits. Monochrome and color (RGB) are more efficient and feasible for detecting outer surface defects, while UV and N cameras are found efficient in detecting defects underneath the peel of the frui[s.
5 1J1U I !II 1i1! 1II IM IIUIII
~a : .r' t F5 '
F is ~ i
f, 1 STUDIESON DESTRUCTIVE AND NON-DESTRUCTIVE QUALITY EVALUATIONS OF MANGO FRUITS
UBMITTED R'.THE-AAWARD OF TFfE_DEGREE OF 'D t!oopjp
r
POST .HA [ VEST EN .INEE IN ANJ+±TECHNOLOGY J:
~k T~1 ~ ,r • 1 mot ; ~`r' _ __ B '1 F -- ,`..} I. _H A ,I XJ M A PAT I-{
1llnder the Super-vis!on of , Dr. Mohammad All Khan - Dr. Abhijit far Department of Post Harvest Division of Food Science & Engineering & Technology Post Harvest Te hnology Aligarh Muslim University, Aligarh IARI, New Delhi - 110 012 r
DEPARTMENT OF POST HARVEST ENGINEERING AND TECHNOLOGY FACULTY F AGRICULTURAL SCIENCES ALIGARH MUSLIM UNIVERSITY { ALIGARHH (INDIA) A pril 2014 . &partment ofcPost J(a st Engineenng a ec noiigy * E Faculty of Agricultural Scien ce Aligarh Muslim University, Aligarh- 202 002, INDIA
GYE~~FIC~2~E'
This is to certify that the thesis entitled "Studies on Destructive and Non- destructive Quality Evaluations of, Mango _fruits" being submitted by Er. Krishna Kumar Patel to Aligarh Muslim University, Aligarh for the award of the degree of Doctor of Philosophy .in Post Harvest Engineeringand Technology, is a record of bonafide research work carried out by him. He has worked under our guidaii a and supervision and fulfilled: the' requirement for the submission of the thesis. The results contained in the thesis have not been submitted in part or full to any other University or Institute for award of any degree or diploma
(Mohammad Ali Khan ) (Abhijit Isar) Supervisor & Chairman Co-supervisor & Senior Scientist, Dept. of Post Har est En . &.Technology, D. of Fzad Sc enee & Post H r+ret'L'c lflola y, F!O A r`icultural Sciences, IARi, New Delhi 110 012 AMU, Aligarh ^ 202 002 Til
n6lf
it ljl~ Vitae
The author was born in district Pratapgarh, in Uttar Pradesh Province of India on May 10. 1979. In 2005, he received his Bachelor of Technology in Agricultural Engineering from the Chandra Shekhar Azad University of Agriculture and Technology, Kanpur, India. He received his Master of Technology degree in Agricultural Process & Food Engineering from Aligarh Muslim University, Aligarh, India in 2007. His dissertation was on "Studies on development, quality evaluations and storage stability of snack food prepared from multipurpose flour (Rice, Wheat & Gram) and carrot powder". He joined Indian Agricultural Research Institute, New Delhi as research associate in the World Bank funded National Agricultural Innovation Sub-Project, on "Development of non-destructive system for evaluation of microbial and physico-chemical quality parameters of mango in 2008. Since then, he has been engaged in research. In 2012 he joined All India Coordinated Research Project on Post Harvest Technology, AMU, Aligarh as Scientist (As & PE). Ho has published seven research papers in national and international journals. He has also contributed three chapters in knowledge based book edition published by renowned publisher. He has also attended nine conferences and has presented papers therein. He is a member of Indian Society of Agricultural Engineers and participated as reviewer in three renowned international journals. His research is currently focused on non-destructive quality evaluation of agricultural produce based on computer vision technology.
List of publications and conference papers on the subject
Paper Published: 03 1. Patel IKC, Kar A, Tha SN, Khan MA (2012) Machine vision system: a tool for on-line sorting and grading of agricultural foods and produces. J Food Sci Technol (March— April 2012) 49(2):12---141
2. Patel K K, Kar A, Khan MA (012), Nondestructive Food Quality Evaluation Techniques. Principle and Potential, Agricultural Engineering Today, Vol. 36(4): 29-
3. Patel KK, Khan MA, Kor A (2012) Recent developments in applications of 1 RJ techniques for foods and agricultural produce - an overview. J Food Sci Technol (1)01: 10.1007/s13 197-012-0917-3)
Paper Published in International Conference: 01 1 Kiir A, Samuel DVK,, Patel 1(IX, Nath A. Mishra B. (2011) Variation in Physico- hemical Properties of Some Varieties of Indian Mangoes. An ASABE Meeting Presentation, August 7 --10, 2011, Paper number 11.
Papers communicated: 03 1 Patel KK, Kar A, Khan MA Kumar Y, Bal LM, Sharma IX (2014) Image processing tools and techniques used in computer vision for quality assessment of food products. 1UF S-2014-104 (A epted). 2 Kar A, PatCI JK, Nath A, Mishra 1 , Jha SN, Khan MA (2014). Modeling of peel color of mango (cv, rliphon3o) fruit for evaluation of harvest qualities, LWT - Food Science aid Technology, L TT- -14-00 58.
Paper Contributed to the ScnihiariConfrrenceiSymposi.: 02 1 Patel I<, Kar A, Khan MA (2012) Quality Assessment of Mango (Mang era indka, cv. Chausa) Using image Processing Techniques. International conference on Population Dynamism and Sustainable Resource Development, Mar+c]i 25-27, paper no 441,2321, 2 K.ar A, I .K. Patel, Mishra R, Nath A (2012). Defect detection on the surface of mango fruits (Mangi era indica cv. Chausa) using image processing techniques, 46 Annual convention of Indian Society of t gxicultural engineers (I AE) and Internatiion symposium on grain storage. Feb 2.7 29 at College of Technology. G.B. Pant University of Agriculture and Technology, Pantnagar, Uttarakhand, Paper ID No. HCP-2012-1-03 Paper to be cummunleated 1. Detection of mango fruits defects using UV based computer vision technique 2. Monochrome computer vision based non-destructive physical characterization of
iv mango fruits. 3. Devc1opin nt of computer vision system for sorting and grading of mangoes. 4. Studies on NIR camera based computer vision technique and its potential for detecting defects just below the peel of mango. 5. Comparative study of monochrome and computer vision system for detecting defects on the external dace of mangoes. 6. Effects of various factors on physico-.chemical properties of some commercial varieties of Indian mangoes. upraise and jrttitu4 to Cmrd of Vniverse (o4 the ta'wst &rnficent and the weft f of Petting m complete this wot 5{e gave ire the courage and patience to 'errome the odls It came trey way during the gc.tatin periado tkis tf~esiK It rs u 6rCssings, w/ i i Free sustained me a[ through this time. ! ? rE rfd not hiave been nEat I am totfay, Ea1 tIe great %rte 9dusim Vnvcersity not graxit we admission to its haf i W Aorta& I wish to pay myfounegrntitude to the founder of c(th cradle o &n&wfedge SirSyed Emc 7 at7 wEm a farsightedness Eas fieed marry a pei ort ij me.
I wish to mess my rr° ourtd jratuti44, respect and hciu>ur io i + vet~ 6fe siw re isar, tDr 911of amrx t l air, C'iairman, DTai ment of Tost- -{arvesr fErI Ueer1!Ig rtad Tecnnn y, !racu t of gri uCturg.( Science. )I.h .' ., I r rt and Co-sapervi.sor Or. , 6Ijtt ?Cti 5 nior . cienti.U, Poa .7{ariest ngiiee ng Division, Ind ai Agnri1turai 5 s rcf Institute, W v i-12 for their ent snit, if a#.ive and p cxaus guidance, cvaesra t supmv& ort r teca( opir tnd dm Cy suygestiorr, constant useful encouragemen t and tech ri a[ tips which has arways been a source o f inspiration during the course of a renmentatio aii preparah oa of this tEesis. ' t' t Eis iris-sterft help and nce, I -rvortfdfum never beeii aG[e to comp1 to the report.
I record my sincere gratitude to my respected teucfiners tEr. .A Shay Stastave~, -Assistant t ro essor aad DDr Sag&r, fi ad, ssitt. JSOI. and seniors +Tf L Iu is acrd Dr. a jabeen Si Egni; Scientist M bit (W1 Jpniors epartmeat of Pos +est LEt rneensg i u Tock iafo~y, +Facdrty of 4ricu ura( Sci'erucs, fliUJ,, Uzth, for lts consttzntconstant ca--Iperatkn. sympathy, sncourge eTil, siwjeston fin ii per ld1*Merest an provlding necessd?y facifltks d ring tkee entire • course o st d5.
I uvuff e to c pn s again my hearcfe1 greti#r.ide ant curter thanksj to the ;Dr,46[r t Kjzç who is my Co- perrrirar, Or. .'[ , Samuel i-iiid Division of (P= Hth'est !E,Pgureenng, Wr Isfam fan, lecf nica(of 5cer an6iITmy team wemers of tEe ietioiaguftnra11rzmwation rojeci,fr his va1uaFie support arc enc i ent.
,Lfi specia(tIl oMkJgo Jo print ti(sthnti t, smior srien tests, scientist. aruf ri n(s in CDivi i of Est .Ffar ' st Engiheering, I.)4.1., yewDeGhfor ter best supports and gu id race. I wwufdti a to t1an al` my team : met,, Efd` iswestigutors of )1 India Co-or4nated Project on mast EMeit 'Tecthiofvgy anir,-!&satww1 h [fan ()for their best co-operatives that cannot 6e neg1eae1drinig the Cetiori of my psis Wt .
vi I wi(1 Se i% to my 1itii s-f{I Hass to express my prof ouni danc(deepest sense of gathwie to myparent -a other aruity m m fort their inter st in my stud' . maxi o(dassistarice, imrmnse srippore and icoitragement, •witGout it was impassi6 to compfete t s fissenation.
nandat su port unier the ' I or Oair&fa ded ativna1,Agrfcuftura( Innoiatior+ .Sub- ject, ors "~D a vpm rtt of tnoai-dstn.cive system emfuation of mic ia(ar pfry rc-d icn quzfsLy parameters of mango being implemented at the Indian Agricufturar esearci I~rstitute is dufy
I wauWuot have beery what I am today. hacf xt not been for the gr?at sacrifices of dry be w( mother anu atI er, -wIosr 5Cwsmgs, constant u iration and ur1strnteisupport are a(uays with me. I am et if oft edta them
# iiou/iRkj to tfaifs m- aIT relatr'ves andmv 1i tfe nepl etv ', ya' affil ' idNagva or tfierr cvnsfant support and encour9gement odor being so ua ersianding Lun rig ny entire reseait!.
Last but not the feast, I wi"sfi to e my gratitude to of the dh'idua& who (ia a direct y or irtdirect4 c tn reévl airy manner in the co ¢p etim! of tfiis -woriE, 2 arr s are due to all the a'ztors (aiid rpublishers) whose wor ,has 6een cftedint ist!esis:
Place: Aligarh (Krishna Kumar Patel)
Date:6I(O51.o1CJ Ph.D (PIlE & Tech.) En.No. B: 9083
vu TABLE OF CONTENTS kapter Title Page No. Cert/icate i Dedication ii Vitae iii List of 'publications and conference papers on the subject iv-v
Acknowledgement V1-Vii
Table of Contents vlll-7L11 List o Tables xiii-xvi List of Figures xvii- -ii Nomenclature and abbreviations xxiii-xxiv INT1ODUCT1ON 1-5 1.1. Introduction and backgrouwd of the present work 1 1.2. Objectives of the study 4 1.3. Thesis purview/ organization 4 LITERATURE REVIEW 6 2.1. Pos arrest operadons and quality management of mango 6 fruits 2.1.1. Postharvest management before marketing 7 2.1,1.1. Pre-cooling R 2.1.1.2. Grading 10 2.1 ,2 Postharvest disease of mangoes 11 2.12J. Anthrjcnose (Mick lesion) 11 2.1..2. Later burn (staining) 12 2.1.2.3. Stem end rot 13 2.1.2.4. chilling injury 14 2.1.2.5. Alternarta rot 15 2.1,3. Packaging 16 2.1.4. Storage 17
2,1,5. Transportation 19 2.2. Principle and Potential of non-destractive food quality 20 evaluation techniques - 2.2.1. X-ray and computed tomography 23 2.2.2. Magnetic resonance imaging (M 1) 24 2.2.3. Infrared 26 2,2.4. Ultrasound 28
VIII 2.2.5. Electronic nose (E ) 28 2.2.6. Computer vision system 29 .3.i chifte/compufer vision system and its principle components 33 2.3.1. Image acquisition system 35 2.3.1.1. Ultra violet imaging system 35 2.3.1.2_ Visible imaging system 37 23.1.3. MR imaging system 3 2.3.1.4. Thermal Imaging 40 2.4, Image processing tools and reehni'q es used in computer 43 vision stern for, food quality evaluation 24.1. Image processing 43 2.4.2. Image processing techniques 44 142.1. Low levo1 image techniques 47 2.4.2.2. Mid-level processing- Segmentation techniques 52 2.4.2.3 High-level processing: image analysis 61 2.4.2.4.Other techniques 62 2.4.3. Applications of image processing tools and techniques 66 used in computer vision system for quality evaluation of food and agricultural products 2.4.3.1, Assessment of fruits 66 2A.3.2. Assessment of vegetables 69 2.4.3.3. For grain classification and quality evaluation 71 2.4.3.4. Processed food products 73 2.4.3.5. Detection and localization of fruits on a trcelplant 77 2.4.3.6. Object measurements and classification 79 2.4.4. Advantages and disadvantages 36 3 MATERIALS AND METHODS 88 3.l.Sample Collection 88 3.2.Physical and chemical analysis 89 3,2.1. Determination of physical properties 89 3.2.1.1. Length, Breadth, Thickness 89 3,2.1.2. Sphericity and Aspect ratio 89 3.2.1.3. Mass, volume and specific gravity 89 3.2.2. Determination of biochemical properties 90 3.2.2.1. Total soluble solids 90 3.2.2.2 .Tit1-atable Acidity 90
ix 3.2.2.3. Total cirotdnoid3 91 3.3 .Non-destructive Analysis 91 3.3.1. Computer Features 92 3.311. Hardware 92 3.3.1.2. Software 92 3.3.2. Image acquisition/ vision system features 93 3.3.2.1. Illumination Chamber 93 3.3.2.2. Cameras and Frame Grabber 94 3.3.3. Processing of images for physical characterization 95 3.3.3.1. Monochrome image processing 96 3.3 ,3.. Color image processing 98 3.3.3.3. Ultraviolet image processing 106 3.3 ,3.4. Processing of NTR image 110 3.3.4. Particle analysis 114 3.3.4.1. Area 114 3.3.4.2. Perimeter 114 3.3.4.3. Max Feret Diameter (MFD) 115 3.3.4.4. Waddel Disc Diameter (WDD) 1I5 3r3,4.5. Elongation Factor 115 3.3.4.6. Compactness Factor 115
3.3.4.7. kiey .00d Circulatory Factor 115 3.3.4.8. Type Factor 116 3.4 image processing for defect dote tioon 116 3.5.bata analysis 122 3.5.1. Geometrical attribute based modelling of fruit weight and 123 volume 3.6.2 Performance evaluation of algorithms 124 RESULTS AND DISCUSSION 125 41. Studies on variability of physico-eheinka! parameters of 125 nar go fruit 4.1.1. Dimensions of fruit 125 4.I .2. AMID and G D or mango size 128 4.1.3. Fruit weight and volume 129 4.1.4. Sphericity 131 415. Aspact ratio 132
x 4.1.6, Specific gravity 133 4.13. Total soluble solids 13 4.1.S. Titratable acidity 136 4.1.9. To 1 carotenoids 137 4.1.14. Color values 138 4.2. Nnn-destructive quality evaluations of mango fruits 141 4..-1. Noi-destrue:ive gnallr' evaluations of mango fruits 141 using inonociwome complster vision technology 4.2.1.1. Size measurements 141 42.1.2. Estimation of shape attributes by image analysis 148 4.213. Relationship between fruit's shape features and 151 weight 4.2.1.4. Relationship between fruit's shape features and 152 volume 4.2.1,5. Geometrical attribute based modeling of fruit 154 weight and volume 42.1.6. I etectiou of external defects of mango fruits using 159 monochrome computer vision technique 4.2.2. Non-destmet ve quality evaluations of mango fruits 15 rising color RGR) computer vision tecu w10 4.2.2.1. Size measurements 165 4.2.2.2. Estimation of shape attributes by image analysis 171 4.2.2.3. Relationship between fruit's shape features and 172 weight 4.2.2.4. Relationship between fruit's shape features and 174 volume 4.2.2.5. Geometrical attribute based - modelling of fruit 177 weight and volume 4.2,2,6. Common external defect detection of mangoes 181 4.2.3. Nan-destructive quality evahual ons of mango fruits using 189 F' computer vision technology 4.2.3.1. Size measurements 189 4.2.3.2. Estimation of shape attributes by image analysis 195 4.2.3.3. Relationship between fruit's shape features and weight t96 4.23,4. L olationship between fruit's shape features and volume 198 4.2.3.5. Geometrical attribute based modelling of fruit weight 202 and volume 4.2.3.6. Detection of mango fruits defects using liv computer 204 vision technique
xi 424 Non-deiructive quality evaluations of mangofruits using 213 NIA computer vision tee! ndo y 4.2.4.1. Size measurements 213 4.2.4.2. Estimation of shape attributes by image analysis 219 4.2.4.3. Relationship between fruit's shape features and weight 220 4.2,4.4. Relationship between fruit's shape features and volume 222 4.2.4.5. Geometrical attribute based modeIing of fruit weight 2.24 4.2,4.6. Defect detection of mango fruits using NIR computer 227 vision technique 5 SIJMILIRY, CONCLUSION, RECOMMENDATIONS AND 233 FUTURE WORK 5.1. Summary 233 5.2. Conclusion 237 5.3. Recommendations for Future work 238 Rcferentes 240 Appendices 276 Appendix-I 276 Appendix-2a Appendix-2b 287 Appendix-2e 289 Appendix-2d 291 Appendix-3 293
xii LIST OF TABLES
Table No. C ption Page No. 2.1 Area, production and productivity of mango in some popular 7 state of India in the year 2010-11 2.2 Nutritional Value of 100 g fresh mango pulp S 2.3 Machine vision versus human vision: capabilities and 22 performance evaluation 2.4 Different components of qualities for fruits and vegetables 23 25 Summnar of machinelcnrnputer vision applications for 32 different areas ' 2,6 Commonly used image processing terms 4546 17 First order statistics which are based on statistical properties 65 of the nay level histogram of an image region 2.8 Summary of machine vision applications for food and 75 agrieultural produce. 2.9 acne fits and drawbacks of ranchinelcomputer vision system 87 3.1 Details of different cameras used for image acquisition of 95 mango its 3.2 Processing steps used in processing of images for 117 segmentation of defects 4.1.1 Effect of harvest stage and cuItivar on some important 126 physical parameters of mango fruit 4.1.2 Combined effect of Factors viz. harvest stage and cultivar, on 127 some important physical parameters of mango fruit 4.1.3 Summary of effect of harvest stage, cultivar and storage 130 period on T S, titratable acidity, total carotenoids, specific gravity and colour values (L, a & b) of mango fruits 4.1.4 Summary of effect of interaction between harvest stage (HS) 132 and cultivar (V) on TSS, iitratable acidity, total carotenoids, specific gratify # id colour values (L a &G ) of mango fruits 4.1.5 Summary of effect of interaction between storage period (SP) 134 and cultivar (TV) on T S, titrrtable acidity total carotenoids, specific gravity and colour values (L, a &. b) of mango fruits 4,1.6 Summary of effect of interaction between harvest stage and 140 storage period (SF) on TSS. titxatable acidity, total carotenoids, specific gravity and colour values (L, a & b) of mango fruits 4..1.1a Summary statistics of the measured physical properties 142 (length, width and thickness) of mango fruit's using calliper and image proce.sing technique based on monochrome CVs.
4..1,1 b Comparison of measured (VC) and computed (IP) length, 144 breadth and thickness of mango fruits used in the analysis 4.2.1.2 Paired t-test analysis, for fruit's length, breadth and thickness 144 of Ciarsa and .Dushchari cultivar of mango, between two measurement methods [vernier caliper and computed image processing (IP)] 4.2,1.3 The mean values, standard deviation (S.D.) and coefficient of 149 shape geometric attributes of mango fruits (Chausa and Dashehari cultivar) analyzed by IF technique based on monochrome CVS. 42.1.4 Summary of coefficient of determination W), SE and linear 150 iegmssiori weight models (forChausa and Datheharr mango cultivars) on the basis of shape attributes obtained by image processing based on monochrome CVS 4.2.1.5 Summary of coefficient of determination (R2), RSE and linear 154 regression volume models (for Chausa and Dashehari mango cultivars) on the basis of shape attributes obtained by image processing based on monochrome CVS 41.1,6 Various MLR models based on IF dimensional characteristics 157 (length breadth and thickness) for determining weight of mango fruits of Chausa and Dashehari cuitivars 4.2.1.7 Various MLR models based on IP dimensional characteristics 1 5S (length, breadth and thickness) for determining volume of mango fruits of Chausa and Dasheharl cultivars 4.2.1.8 Performance of proposed algorithm 1 64 4.2.2.1 a Summary statistics of the measured physical properties 166 (length, width and thickness) of mango fruit's using caliper IP method ba.sed on color CVS. 4.2.2.16 Comparison of measured (VC) and computed SIP) length, 167 breadth and thickness of mango Fruits used in the -analysis 42.2.2 Paired t-test analysis, for fruit's length, breadth and thicimess 167 between two msasurement methods [vernier caliper and computed by image processing] 4.2.2.3 The mean values, standard deviation (.D.) and coefficient of 171 shape geometric attributes of mango fruits (Chausa and Dashehari cultivar analyzed by IP technique based color . 4.2.2.4 Summary of coefficient of determination R2), RSE and linear 172 regression weight models (for Chausa and Oashehari mango cultivars) on the basis of shape attributes obtained by image processing based on color CVS. 4.2.2.5 Srnmmary of coeff icient of determination '), RS and linear 176 regression volume (y) models (for ChQusa and Dashehar, mango cultivars) on the basis of shape attributes obtained by
xiv image processing based on color CVS. 4.2.2.6a. Various MLR models based on III dimensional characteristics 178 (length, bleadth and thieknes) iDr [ termiuing weight of mango fruits of Chausa and Dasiehari cultivars 4.2.2.6b Various MLR models based on IP dimensional cha cteristics ISO (length, breadth and thickness) for determining volume of mango fruits of Chm.sa end Da..hehari cullivars 4...7a Calculated efficiency of various defect types 188 4..2,?b Calculatcd accuracy of various defect types 189 4.2.3 A a Summary statistics of the measured values of mange fruit's 190 Length, Breadth and Thickness when UV camera based CVS was used. 4.2.3.1b Comparison of measured (V) and computed (IP) length, 191 breadth and thickness of mango fruits used in the analysis 4.2.3.2 Paired t-test analysis, for fiuit's length, breadth and thickness 191 of Chausa mango cultivar, between two measurement methods [vernier caliper and computed (IP)] 4.2.3.3 The mean values, standard deviation (S.D.) and coefficient of 196 shape geometric attributes of mango fruits (Chausa and Dushe!+ari c,nitivav) analyzed by IP technique based on UV 5. 41,3.4 Coefficient of determination (R2) and linear regression weight 198 models on the basis of area (pixels) obtained by image processing technique based on UV . 41.3,5 Coe#licient of determination 2) and linear regression 200 volume models on the basis of area (pixels) obtained by image processing technique based on UV CVS. 4..3.6a 'Various MLR models based on IP dirmensional characteristics 201 (length, breadth and thickness) for determining weight of mango fruits of Chsa and Dashehari cultivars 42.3.6b Various MLR models based on IP dimensional characteristics 203 (length, breadth and thickness) for determining volume of mango fruits of Chausa and Z)ashehari cultivars, 4.2.3.7 Various defects and their appearance in UV and color images. 212 4..4.Ia Summary statistics of the measured values of mango fruit's 214 Lend, Breadth and Thickness when NIR CVS was used. 42.4. lb. Comparison of measured (V) and computed (IP) length. 215 breadth and thickness of mango fruits used in the analysis, 4.2.3.2 Paired t-test analysis, for fruit's length, breadth and thickness 215 of Chausa mango cultivar, between two measurement methods [vernier calliper and computed (1P) . 4.2.4.3. The mean values, standard deviation (S.D.) and coefficient of 219 shape geometric attributes of mango fruits ( lu sa and
XV Dashehari cultivar analyzed by 1P technique when NIR CVS was used 4.2.4.4. Coefficient of determination (R2) and linear regression weight 220 models on the basis of area (pixels) obtained by image processing technique based NIR CVS 41.4.5. Coe.fficcnt of drt ination (R2 and linear regression 222 volume models on the basis of area (pixels) obtained by image p ucessing technique based on NIR C.VS. 4,2.4.6a Various MLR models based on IP dimensional characteristics 225 (engt .breadth and thickness) for determining weight of mango fruits of Chausa and Dus xehar cultivars. _ 4;23.6b, Various MLR models based on IP dimensional characteristics 226 (length, breadth and thickness) for determining volume of mango fruits of Oiausa and Daslie/rari cultivars.
xvi
LIST OF FIGURES
Fig. Caption PAGE: No. No. 2.1a Diagram of typical computer vision system 21 2.1 b Human vision system 21 2.2 Block diagram of an MRI system. 25 2.3 Schematic diagram of the experimental setup for NIR testing of peach 27 frail 14 Image of the portable EIectronic Nose developed at Th M-CNR — Lecce, 29 Italy 2.5a Principle components of machine vision system 34 2.5b Functional block diagram of basic machine vision system 34 2.6a Configuration of common image processing system. 44 2.6b Various steps of low-, mid- and high-level image processing techniques 48 2.7 Typical segmentation techniques: (a) thresholding based segmentation ib 55 edge based se entatior+ (c) region-based segmentation 2+8 Pizza images: (a) original image; (b) segmented image 57 2.9 Determination of Feret's diameter 81 3.1 Schematic diagram of computer vision system 92 3. 2 Images of mango fruit acquired at different postures and respective 93 segmented and equalized image
3.3 A glimpse of computer visioalimaging system used in this research 94
3.4 A glimpse of different cameras (A) IJV Camera; (B) Visible monochrome 95 camera; and (C) NIR Camera
3.5 Shows histogram of original image (I), image after improvement of 96 brightness ), and. of image after HIS-intensity plane extraction (3) for Chawsa and Dashehari fruits in case of monochrome camera. 3.6 Flow diagram of algorithm for dimension and other particle 99 _measurements of Chausa fruits using monochrome CVS (a) Original monochrome image, (b) image after brightness enhancement, (c) image after HIS-intensity plane extraction, (d) image after threshold, (e) filtered image, ( the outline of image, and (g) image with length and breadth measurements. 3.7 ' Flaw diagram of algorithm for dimension and other particle 100 measurements of Dashehari fruits using monochrome CVS (a) Original monochrome image, (b) image after bright css enhancement, (c) image after hI -Intensity plane extraction, (d) image after threshold, Ce) filtcrcd image, ( the outline of image, and (g) image with length and breadth measurements. 3.8 Some geometrical attributes of mangy} fruit 101 3.9 Shows histogram of original image ( ,& A2), histogram of image after 102 improvement of brightness (Bt& B,), and histogram of image after blue- RGB color plane extraction (C i A C2) for Chausa fruits in case of color (ROB) camera 3.10 Shows histogram of original image (.A1& A2) histogram of image after 103 improvement of brightness (B[& B2). and histogram of image after blue- ROB color plane extraction (C[& C2) for Dasht rari fruits in case of color (RUB) camera 311 Flow diagram of algorithm for dimension and other particle 104 measurements of Chausa fruits using color (ROB) CYS. (a) Original ROB image, (b) image after brightness enhancement, (c) image after blue plane extraction, (d) image after threshold, (e) filtered image, f) the outline of image, and (g) image with length and breadth measurements 3.12 Flow diagram of algorithm for dimension and other particle 105 measurements of Dashehari fruits using color (RUB) CVS. (a) Original ROB image, ,b) image after brightness enl ncement, (c) image after blue plane extraction, (d) image after threshold, (e) filtered image, ( the outline of image, and (g) image with length and breadth measurements 3.13 Shows histogram of (1) original image, (2) image after improvement of 106 brightness, and (3) image after iutenaity-HSI plane extraction of Chaa (A) and Dashehari (B) fruits in case of IJV camera 3.14 Flow diagram of algorithm for dimension and other particle I08 measurements of Chausa fruits using UV CVS. (a) Original UV image, (b) image after brightness enhancement, (c) image after HSI-Intensity Plane extraction, (d) image after threshold, (e) filtered,. image, (t) the outline of image, and (g) image with length and breadth measurements 3.15 Flow diagram of algorithm developed for dimension and other particle 109 measurements of Dashehari fruits using UV cvs. (a) Original UV- image, Ib) image after brightness ernhancem nt, (c) image after MS- intensity plane extraction, (d) image after threshold, (e) filtered image, (t) the outline of image, and (g} image with length and breadth measurements 3.16 Shows histogram of (1) original image, (2) image after improvement o# 11-0
Xviff
brightness, and (3) image after intensity-H I plane extraction of Chausa (A) and DasJieha - (B) fruits in case of NIR CYS 3.17 Flaw diagram of algorithm for dimension and for other particle 111 measurements of Chausa fruits using NIR CATS, (a) Original NUt image, (b) mage after brightness nhancement, (c) imagc after H l-Xntensity Plane extraction, d) image alkr threshold, (e) filtered image, (fl the outline of image, and (g) image with length and breadth measurements 3.18 Flow diagram of algorithm developed for dimension and other particle 112 measurements of Dashehari fruits using N1R. . (a) Original I TIR image, (b) image after brightness enhancement, (c) image after HSI- Intensity Plane extraction, (d) image after threshold, (e) filtered image, (1} the outline of image, and (g) image with length and breadth measurements 3,19. Algorithm steps for defect detection on surface of Chzeaa and Dashehari I 1 S cultivars by using monochrome CVS 3.20, Flow diagram of developed algorithm for defect detection on the surface 114- of f nits of both cultivars (Chausa and Dashehari) using color (ROB) 120 Cvs. 3.21. Flow diagram of algorithm steps for defect detection underneath the 121 fruit's surfncrcipeel using NIR CVS 4.2.1.1. Comparison of computed and measured outer dimensions (L. W, T) of 146 'hausa (a, b, c) and Dashehari (d, e, #) fruits with image processing (IP) method based on monochrome CVS and digital vernier caliper (DVC). 4.2.1.x, Bland—Altmmn plot for the comparison of outer dimeusions (L, W, T) of 147 C/iausa (a, h, c) and Dashehari (d, e, fhuit's computed with image processing (I) method based on monochrome CVS and measured with vernier caliper method ( M); outer lines indicate the, 95% limits of agreement and center line shows the average difference. 4.2.1.3. Correlations between shape attributes tarea (a1, a2 , perimeter (b1, b,), 153 Max Feret's (c,, c•) and Waddel Disk (d', d2) diameters] of Chausa (1) and Dashehari ( fruits computed with particle analysis by image processing (IP) method based on monochrome CVS and weight of fruits measured by digital electronic balance (DEB'.
4.2.1.4. Correlations between shape attributes tarea (a,, a2), perimeter (b1, b2., 155 Max Feret's (CL, c2) and Waddel Disk d1, d diameters] of chausa (1) and Dathehari (2) fruits computed with particle analysis by image processing (JP) method based on monochrome C S and volume of fruits measured by water displacement method. 4.2.1.5. Images of severally damaged (10-49%) fruits with anthracnose disease 161 captured by monochrome camera and of related defects on their surface
Xix er segniezihtion. 4.2.1.6. Monochrome image of slight diseased fzuits with infected area 1-9% (1 to 162 4 and 9 to 12) and image after defect segmentation (5 to S and 13 to 1 ) on their surface. 4.2.1.7. Monochrome images of mechanically damaged fruits (1, 2 & 3) and 163 segmented defects (4, 5 &.6) on their surface. 4,2.1.8, Images of shrivelled fruits with various damages (1, 2, 3 and 7, 8, 9) 163 captured by monochrome camera and segmented defects (4, 5, 6 and 10, 11, 12) on their surface. 4,2.2.1 Comparison of computed and measured outer dimensions L W. T) of 163 Chausa (a, b, c) and bashehari (d, e, f fruits with image processing ([P) method based on color CVS and digital vernier calliper (DVC). 4.2.2.2 Bland—Altman plot for the comparison of outer dimensions (L, W, T) of 170 Chao, a (a, b, c) and Dasheharr (d, e, f) fruit's computed with image processing (JP) method based on color CVS and measured with vernier calliper method 1V CM); outer lines indicate the 95% limits of agreement and center line shows the average difference.
4.2.2.3. Correlations between shape attributes [area (a1, a2), perimeter (1 L, b2), 173 Max Feret's (c1, c2) and Waddci Disk (di, d2) diameters] of Chausa (1) and .ashehuri () fruits computed with particle analysis by image proccssirrg (IP) method based on color CVS and weight of fruits measured by digital electronic balance (DEB). 4.2.2.4. Correlations between shape attributes [area %, az), perimeter (ray, b2), 175 Max Feret's (ci, C2} and Waddel Disk (d1, d2) diameters] of Chausa (1) and l)ashehar i (2) fruits computed with particle analysis by image processing (IP) method based on color CVS and volume of fruits measured by water displacement method 4.2.2.5 The images of fresh mango fruits of Chausa cultivar with (0-1%) or 183 without defects (A), hue plane of color CB) and results of defects (C), segmented using proposed method. 42.2.6 The images of fruits with severe black lesions (anthraenose) and results of 184 proposed algorithm after defect segmentation. 4.2.2.7 The images of mango fruits with severe black spots because of physical 184 or pathological or insect damage and results of proposed aigorithrn after defect segmentation. 42.2.8 (A) Various images of mango fruits with different common minor defects 186 (6-9%) on skin in different pasture, and (B) results of proposed algorithm after defects segmentation. The bright spots (B) on surface are related defects.
xx
4.2.2.9 The images of shriveled mango fruits with black lesions and results of 187 defects segmented using proposed method. 4.2.31. Comparison of computed and measured outer dimensions (L. WI, T) of 193 Chausa (a, b, c) and Dashehari (d e, 'Tifruits with image processing tIP) method based on UY CVSand digital vernier caliper (DVC). 4.2.3.2 Bland—Altman plot for the comparison of outer dimensions (L, W, T) of 194 ChQusa (a, b, c) and Dashehari (d, e, fruit's computed with image processing II') method based on UV CYS. and measured with vernier caliper method (VCM); outer lines indicate the 95% limits of agreement and center Iine shows the average difference. 4.2.33. Correlations between shape attributes [area (as, a2, perimeter (bi, ba), 197 Max Feret's (ci, c2 ) and Waddel Disk (di, d2) diameters] of Chaso (1) and Dashehari (2) fruits computed with particle analysis by image processing (IP) method UV CYS and weight of fruits measured by digital electronic balance (DEB). 4.2.3.4. Correlations between shape attributes [area (ai, a2), perimeter (b1, b2), 199 Max Feret's (ci, c2) and Waddel Disk (di; d diameters] of Chausi (1 ) and Dashehari (2) fruits computed with particle analysis by imams processing (lP) method UV CATS and volume of fruits measured by vaster displacement method 4.2.3.5. Appearance of mechanically damaged areas on the surface of mango fruit 205 in the image captured by UV and color cameras 4.2.3.6. Appearance of severely defected fruits with anthracnuse (Black lesion) 207 and stern-end rot in color and LTA image, 4.2.3.7. Appearance of Anthracnose (BIack lesion) and fungally decayed mango 208 fruits in UV and color image 4.2.3.8. Appearance of various shrivelled mango fruits in UV, color and cut away 209 image 4.2.3,9. Appearance latex stains on fruits surfaces in color and UV images 211 4.2,4.1. Comparison of computed and measured outer dimensions (L, VV, T) of 217 Chausa (a, b, e) and Dashehari (, e, ) fruits with image processing (IP) method based on NIR VS and digital vernier calliper (DVC). 4.2.4,2, Bland—Allman plot for the comparison of outer dimensions (L, W, T) of 218 Chausa (a, b, c) and Dashehari (, e, f) fruit's computed with image - processing (IP) method based on NIM CVS and measured with vernier caliper method (V M); outer lines indicate the 95% limits of agreement and center line shows the average difference. 4.2.4.3. Correlations between shape attributes [area (at, a2), perimeter (bi, bz), 221 Max Feret's (ci, Ci) and Waddel Disk (dl, d2) tliametets] of chausa (1) and Da hehari (2) fruits computed with particle analysis by image processing (IP) method based on NIR CVS and weight of fruits measured by digital electronic balance (DEB). 4,2.4.4. Correlations between shape attributes [area (a a.), perimeter (bL, b2), 223 Max ,Feret's (e', c2) and Waddol thsk (d1, d~) diameters] of Chui (1) and Dashehari (2) fruits computed with particle analysis by image processing (I) mat1iod based on NIR CVS and volume of fruits measured by water displacement method. 4.2.4.5. Images of mango fruit acquired by using NIR image acquisition system 229 through different wavelength ('tin) filter 4..4.. Appearance of defected areas on fruit's surface in color image, NIR 231 image and after peeling Nomenclature and Abbreviations
Oc Degree centigrade ADC Atialogue to digital converters AFPM Autonomous Fruit Picking Machine AIM Arithmetic mean diameter ANOVA Analysis of variant AO : Association of O±ficiai Agricultural Chemists BMP Bitmap CCD Charge coupled device CRT Cathode ray tube CSM Charge simtdati:n method CT : Computed tomography Coefficient of variation CWRS Canada Western Real Spring CM3 : Cubic centimeter DEB : Digital electronic balance D : Digital vernier calliper EN : Electronic nose GB : Gigabyte GDP : Gross domestic product GMD Geometric mean diameter Gram HIS Hue-Intensity-Saturation IP Image processing Iii International unit Kg Kilogram Lab VIEW Laboratory Virtual Instrument Engineering Workbench LED Light emitter device LUT Lookup table MFD : Max Feret Diameter (MFD) M : Magnetic resonance imaging r~prn Millimetre square ms Millisecond 11I : National instrument Corporation [-]MA : National instrument for image acquisitioli [R : Near infrared imaging 7 Nanometre D. Optical density Peripheral Component Interconnect Correlation coefficient Aspect ratio % Random access Jnemory Retinal equivalents Radio-frequency 1B Red-Green-Blue 1 : Humidity I : Region of interest Regression Standard error D. : Standard deviation Standard error in correction ?P Standard error in prediction I Sphericity 'SS : Statistical Product and Service Solutions xE Standard error in estimate ~i Titratable acidity F : Tagged image file i Total soluble solids ? Un-interrupted power supply 3A united state of America 3B Universal serial bus V Ultra violet ;ITT Vernier .calliper method DD : Waddel Disc Diameter
xxry Chapter X
Introduction
1.1. Introduction and background of the present work
The Indian economy is poised to achieve a double-digit growth rate. Increasingly, India is being regarded as the economy to watch and various projections suggest that India would be the second largest economy by 2050. However, this success story hides a larger worry on the agricultural front. The shy of agriculture sector in DP has declined from around 35 per cent (in 1990-91) to around 27 per cent (in 1999-2000) and further to merely 1L5 per cent during 2046-07. The aixual average growth rate for the agriculture sector as merely 3 per cent in the first five years of the new rnillenniurn. Thus, the agriculture sector is proving to be a drag on the economy. With more than half the population depending directly on this sector, low agricultural growth has serious implications.
Further, the food grains production is actually declining which can have its own impact on the food security of the country. To secure food for everyone, the revitalization of apiculture sector is needed and for this, the growth of food processing-industry has to be ensured. Since, the food processing industry can address the, key issues of wastages and value addition, and attract new investment in the sector. In addition, global ecperiences indicate that agriculture developmee.t in the country can be given a big boost by the growth of agro aid fwd processing industries.
India currently produces about 50 million tonne of fruits, which is about 9 per cent of the world's production of fruits and 90 million tonne of vegetables, which accounts for 11 per cent of the world's vegetable production. However, only about 2 per cent of the fruits and vegetables in India are processed, which is very low compared to countries like the Philippines, Malaysia, USA, China and others. Lack of processing and inadequate storage facilities, for fruits and vegetables, results in huge wastages. Among the fruits, mango (Mangife' ra indica L.) is the oldest, tropical (Hashirn et al. 2006), most popular (Rajkumar et al. 2007), climacteric (Chien et al. 2007), attractive (Sabato et al, 2009) and considered to be the king of fruits of India (hha at al, 2010; l thorn et al. 2007), Indian subcontinent is I cultivating mango for well over 4000 years and mango is native to Southern Asia, especially Burma and Eastern India. Later on it spread to outside of India, especially to Africa, Brazil,
Caribbean and Central America (lagtiani eta 19), during 62 to 645 AD (ha at al. 2010). The details of the mango cultivars, cultivation and quality have also been mentioned in Ain I-A cbari written by in Ahul k'azl (Bose and Mitra 2)01 .
India ranks first in area and production. of mango with 2.31 million ha and 12.75.. million tonnes, respectively with share of 56 ° production in the world. Today, India is growing nearly 1000 varieties of mango. It has high commercial value in the international fruit market and is liked by people for its rich luscious aromatic flavour and a delicious taste with evenly blended sweetness and acidity. Nutritionally, it is rich source of prebiotic dietary fiber, carotenoids, organic acids, polyphcnols, provitamins. minerals, etc. (Anon 2010).
Though India is the world's largest producer of mangoes, but surprisingly, it accounts for less than one percent of the global trade because of lack of rapid, non-destructive and precision sorting methods for quality and safety assurance and many times our export consignments have been rejected at destination port due to detection of soft tissue and some, bacteria inside mango and over, and under ripeness etc. Thus, to satisfy the increased awareness, sophistication and greater expectation of consumers, it is very necessary to improve q,oality evaluation techniques of food products (Brosnan and Sun, 2QO4).
In addition, the ultimate -objective of production, post harvest processing, handling and distribution of fruits and vegetables is to satisfy the consumer, Consumer satisfaction is related to the quality which can be viewed as an absence of defects and a degree of excellence, The colour, size, shape, mass, volume and surface area, etc. are also important `parmm tens', which are considerable during the sorting of mangoes. Thus, to evaluate the Euit's quality, a number of methods have been developed by the researchers generally depends on quality orientation of any food products. Methods could b divided into two types as: analytical/objective methods and subjective/sensory methods. Both methods have their own advantages and disadvantages. Analytical methods are based mi product attributes, whereas, subjective or sensory methods are consumer oriented, For scientific works we need to measure attributes numerically and thus may not be varied in those perspectives. Sensory 2 attributes changes with sensory panels, place, religion, society, and so on. PractcaI1y sensory method of quality evaluation is better for adoption in that region whereas the objective evaluation may be helpfuI in development of specific instruments for specific quality attributes. The objective methods too are of two types: one cam be said as destructive methods and other nondestructive. Most destructive methods use small samples and utilize them during investigation and the used sample will not be again reusable by the consumers. Generally, destructive methods are based on chemical analysis and used at laboratory Ievel,
One drawback of destructive method is that It is not necessary that whatever attributes you have measured in sample will be closely related with bulk from where samples had been drawn. `Where must be substantial variations, In contrast in nondestructive methods, samples or bulk of materials even remain untouched and samples are not destroyed. It remains intact for consumer and consumer can use it even after testing.
There are several non-destructive techniques that have now been emerged potentially in the field of agriculture and its allied sectors. Some of them are as; computer vision, X-ray and computed tomography, magnetic resonance imaging, near infrared, ultrasound and electronic nose and eye, etc. Since, appearance is the tr$t impression that the consumer receives and the most important component of the acceptance and eventually of the purchase decision. Computer vision one of them (non destructive techniques) has been adopted for quality evaluations of mango rioii-.destructively on the, basis of deferent surface parameters. In spite of this, different studies have indicated that almost 40 % of the consumer's decision regarding the purchase of food products is based on appearance. Shape is one of the subcomponents more easily perceived, although in general, it is not a decisive aspect of quality, except in case of deformations or morphological defects. In some cases, shape is a ripeness index and therefore an indication of flavor (Camels, 2004).
Further, computer vision being an objective, consistent, quantitative, rapid, non- contact and non-destructive quality evaluation tool, has been attracting much research and development attention among the non destructive technologies from the food industries, A rapid development has been increasingly taking, place on lity inspection of a wide range of food products .(Sun, 2004; Timrnermans, 1998). Gunasekaran (1996) reported that by the use of computer vision technology, food industries ranked among the top 10 industries. In 3 spite of above, medical computer vision or medical image processing is one. of the most prominent application fields. This area is characterized by the extraction of information from image data for the purpose of making a medical diagnosis of a patient. GereraIly, image data is in the form of microscopy images. X-ray images, axigiography images, ultrasonic images, and tomography images (Anon 2009). For more details of image processing appHcation, interested readers can refer to papers and books (Gumus et al., 2011; Jackman et al., 2011; Patel at al., 201 lb; Sun, 2008; Eheng et al., 006a,b; Youssef et al., 2005; Du and Sun, 2004a; Chen et al., 2002; Gonzalez and Woods 2002: Zuch 2000; Ballard and Brown, 1982).
So far many researchers have been devoted considerably for effort towards the development of machine vision systems for different aspects of quality evaluation and sorting of agricultural products. As a result, new algorithms and hardware architectures have beer developed for high-speed extraction of features that are related to specific quality factors of fruits and vegetables. A number of sorting systems that use computer-vision techniques for evaluating external quality factors are commercially available.
1.2. Objectives of the study
The rese ` h work of above mentioned title will be governed under following objectives; • To identify specific quality attributes and common external defects, which govern the quality standards and markets prices (destructive & non-destnictive) + To capture image using imaging systems and lighting regimes • To process the captured images and develop algorit} s to correlate the image properties with quality parameters
1.3. Thesis purview/ organization
The thesis consists of five chapters. The first chapter illustrates the status of production of mango fruits in India and the importance of computer vision system among the several non- destructive techniques for food quality evaluation. The motivation for the present work and its objectives are highlighted therein. The second chapter contains the review of the relevant literature in the fields of non-destructive quality evaluations based on computer vision
4 system, An effort has been male to furnish up to date information on computer vision system and post harvest operations arid management of mange fruits. Chapter 3 provides a brief description of the sample collection, the cameras used for image acquisition, the methodology of development of algorithms for physical characterization as well as for defect detection and data analysis procedure. Chapter 4 contains results and discussion which is divided into sections A & B. Section A discussed the results obtained from destructive/ subjective or Iaboratory or traditional methods. The results pertaining to monochrome, color, UV and NlR camera based computer vision system (non-destructive methods) are presented and discussed in the section B. The last chapter contains summary and conclusions drawn from this chapter. Finally, in this chapter, some recommendations have been made for further research iii the light of above work done,
5 Chapter Literature Review
This chapter includes the synthesis and analysis details of various published literatures, research papers, books, articles and other sources such as leaflets, etc. reviewed before the start of research. The literature pertaining to varies aspects relevant to the present study has been reviewed under the fal Lowing headings.
2.1: Postharvest operations and quality management of mango fruits 2.2: Principle and Potential of non-destructive food quality evaluation techniques 2.3: Computer vision system and its principle components 2.4: Image processing tools and techniques used in computer vision system for quality evaluation of food and agricultural produce
21. Postharvest operations and quality management of mango fruits
Mango (Marigr era irrdica L.) is the most important and popular fruit grown in many tropical and sub-tropical region of the world. India is the largest producer of mango fruits covers about 35 percent of area and accounts 22 percent production of total fruits in the. country (Gill et al. 2005). It is mainly cultivated in Uttar Pradesh, Andhra Pradesh, Bihar, Orissa, Ka mataka, Tamil Nadu, (3i arata Kerala, Madhya Pradesh, Maharashtra and Wvat Bengal (Table 2.1). Unfortunately, 25-30 percent of mango produce is lost due to improper post harvest operations; as a result then is considerable gap between the gross production and net availability (Gill et al. 2005). Thus, the application of proper operations and handling practices after harvesting of mango ( crngtera indita L.) omits is very importance to provide outstanding quality of fruits in markets. Like pre-harvest operations for the good quality preducti n of mangoes, post harvest operations are also cause of concern, in the, countries with high production of mango, to maintain the quality parameters such total soluble solids (%rix), maturity, color, size, sweetness, oveniunder ripe, etc as per consumer demands and minimize the losses. In addition, the quality and market value of fruits have been found decayed because of several defects and disease such as Anthracnose (black lesion), Latex bum (staining), Stem end rat, Chilling injury, Alternaria rot, etc. during pre- and-post harvest operations.
[. Indian mangoes (fresh) go to the Bangladesh, Bahrain, Fmcc, Kuwait, Malaysia, Nepal, Singapore, U.Ki and Gulf countries, and efforts have been made to exploit European, American and Asian markets. Besides these, the different products of mango in tuding chutney, pickles, Darn, squash, pulp, juice, nectar, slices, etc. are being exported to U.K., U. ,A., Kuwait and Russia. Thus, tht mango fruit is not only tnn med ripe form but also used to produce mango powder, chutney, athanu, pickles; side dishes, from the greenlraw but mature fruit or may be eaten raw with salt, chili, or soy sauce. A cooling summer drink called panna orpanha is also prepared from mangoes.
Table 2.1: Ares, production and productivity of mango in some popular state of India in the year 2010-11 Name of the tato Ar (Ob0'ha) Production(OUO' tt Productivity(# /ha) Andhra Pradesh 399.3 3194.3 8.00 Uttar Pradesh 251.5 2673.3 10.60 Karnataka 124.5 123&S 9.90 Bihar t 40.2 1222.7 8.70 Gujarat 96 772,1 5.0O Maharashtra 444.5 639.6 1.40 Tamil Nadu 125,1 537.8 4,30 West Bengal 70.1 513.3 7.30 Kerala 88 511.1 5.80 Orissa 125.3 428.8 3,40 Others 15{x,2 809.1 5.20 Total 2020.7 12537.9 (0I, 2013)
Thos, to achieve optimum quality of mango fruits before the exporting the rmango fruits, various operation and treatment should be applied under the scientific vision.
2.1.1. Fastha est management before marketing
Before marketing of the mango fruit some past. harvest activities such as Curing, washing, grading, packaging, storage, transportation, processing and marketing etc are the important past-harvest operation to attain good quality and market value. Curing is necessary
7 when fruits are to be stored for long periods to extend the shelf life. The fruits are usually spread on the floor in the orchard's yard, having the cushion of paddy straw or simple straw for nearly 24 hrs and then washed to remove the dirt.
Table 2.2: Nutritional Value of 100 g fresh mango pulp Constituent Amount in 100 Cinstituet.t Amount in 100 g fresh pulp g fresh pulp water 31.7 g Vitamin A, RE 389 meg_RE Energy 65 kcal (272 kj) Vitamin E 1.120 m ATE Protein 0.51 g Tocopherols, alpha 1.12 mg Fats (.27 g Lipids Carbohydirates 17.00 g Total saturated fatty acids 0.066 g Total dietary fiber 1.8 g Total monouns aturated fatty acids 0.101 g Ash 0.50 Total poly unsaturated fatty acids 0M51 g Mineral Cholesterol 0.00 mg Calcium 10 mg Amino acids hot'. 0.13 rng Tryptephan 0.QOS g Magnisium 9.0 ing Threonine 0.019 g Phosphorus 11 mg Isoleucine 0.01 S g Potassium 156 mg Leucine 0.031 g Sodium 2 mg Lysin 0.041 g zinc 0.04 mg Me#hionine 0.005 g Copper 0.11 mg FhenyIalanine 0.017 g Mangnese 0.027 mg T cosine 0.01 g Selenium 0.6 mcg Valine 0.026 g Vitamins Arginine 0.019 g Vitamin C (total 27,2 jag Histidine 0.012 g ascorbic acid) Thiamine 0.056 mg Alanine 0.051 g Riboflavin 0.57 mg Aspartic acid 0,042 g Niacin 0.584 rng Glutarnic acid 0.06 g Pantothenic acid 0.16 nag Glycine 0.021 g Vitamin B6 0.16 rng Praline 0.1118 g Total folate 14 mcg Serine 0.022 g Vitamin A1 RI 3894 IU (USDA, 2001)
in 21.1i.. Pt e--cooling
Post-harvest cooling or precooling rapidly removes field heat from freshly harvested fruits{commodity which helps in stabilizing the metabolic process and reduce deterioration prior to transportation or storage (Ryall and Lipton 1973, Nowak and Mynett 195, Janick 1986) by suppressing enzymatic degradation and respiratory activity (softening), by slowing or inhibiting water loss (wilting) and growth of decay-producing microorganisms (molds and bacteria), and by reducing production of ethylene (a ripening agent). Waslcar and Dhemre (2005) investigated the effect of pre-cooling on extending the postharvest life of Kiesar mango fruits and reported that the precooled fruits had maximum storage shelf life than the un-precooled. Since, temperature, one of the most important environmental factors, plays an important role in the deterioration of harvested fruits by influencing or regulating the respiration. The every 10° increase in tissue temper2ture causes to raise the rate of respiration about 2 to 3 times (Atkin and Tjuelker, 2003). Thus, precooling is an important postharvest unit operation fo.r tropical fruits like mangoes (R.avindra and Goswanni, 2008) and highly uselEhI in extending the storage shelf life and maintaitiing the qualities thereby getting good market values of produce Herdeuburg et al. 1986). In addition, post-harvest precooling also provides marketing flexibility by making it possible to market at the optimum time. Few of the cooling methods are room cooling, forced air-cooling, hydro cooling ((30I, 2013; Thompson et al. 2002), running water, ice cooling, evaporative cooling, etc (Puttaraju and Reddy, 1997). Effect of precooling on the storage life of mango (cv. Maiiika1) using various precooling methods such as hydro-cooling (15 and 30 min, running water (15 and 30 min) room cooling (30 and 60 min), ice cooling (15 and 30 min) and evaporative cooling was studied by Puttaraju and Reddy (1997). They reported that the precooling of mango fruits, immediately after harvest, delayed ripening without any deterioration in fruit quality. However, all the methods used were not equally effective, for instance, cooling under running cold (4-5)C) water for 30 min effectively lowered the fruit temperature by 169C and significantly retarded the ripening, thereby, extended the storage life by 3 to 4 days, and also retained the fruit quality.
Although, ice coaling was the most effective method in reducing the physiological loss of weight, delaying ripening and maintaining higher fruit firmness, the sensory quality of these fruits was unacceptable, due to high spoilage. Similarly, evaporative and room cooling methods were not effective in enhancing the storage life of mangoes, The room cooling or cold rooms are generally wa'k-in type and avoids the rapid cooling of the produce which is desirable. Even though, it is desirable, the cooling should not result in any chilling injury to the fruits. According to Ravindra and Goswam,, (2008), liquid nitrogen (LN2) is a coaling medium with ample potential for use in precooling operations because of its high cooling capacity and inertness of the vaporized nitrogen gas; however, its severe cold temperatures raise concerns an the possibility of chilling injury to the fruits. They have compared a system where liquid nitrogen (L 2) was employed in combination with mechanical refrigeration system as the coaling medium with the common precooling techniques (hydro-cooling and air cooling) for mangoes (cv. Amrapali) and reported that the LN2 system (LN2 flow rate: 20.5 kg/h; average gas temperature: --85° ) had no adverse effect on the quality of the fruits, and improved the cooling coefficient of the air cooling process by 40%. Hence, judicious design of the system and control of exposure time would help in realizing the potential of LN2 in precooling opermtiorns for fruits, which would be practically useful in technologies like control atmosphere storage system in particular.
2.1.1.2. Grading
It has been observed that the ripening time for bigger size fruits and smaller size fruits is difference. The bigger size fruits usually take 2-4 days more in ripening than'smaller ones and hinder to achieve uniform ripening. Hence, the packing of smaller fruits with larger once should be avoided to achieve uniform ripening. Immature, overripe, damaged and diseased fruits are discarded (Pradeep2 umer, 2008). Shape, weight, colour and maturity are the other important quality parameters for grading of fruits before exporting. For instance, the export quality mangoes are categ The physical properties of mango, in addittou, are important in he designing of equipments for harvesting, sorting, storing, grading, packaging and processing. These properties also affect the handling or conveying characteristics and in the estimation of 10 cooling and heating loads ( ohsenin 1978; Razavi and Parvar 2007), Various studies have been published about the physical, mechanical and biochemical properties of fruits, such as grape berry (Fava at a] 2011), kiwifruit (C mgi at al 2011), peach (Ahmadi et al 2010), pomegranate (Celik and Ercisli 2009), barber (FathoIlah.zadeh and Rajabipour 2008), persimmon IIgual et al 2008), apricot (Ha eferogulSari et al 2007), guna fruit (Aviara et al 2007), aonla (Loyal et al 2007), myrtle (Aydin and Ozcan 2007), fresh oil palm (Owolarafe et at 2007), strawberry (Ozean and Haciseferogullari 2007), tomato (Baltazar at al. 2007), peach, nectari res, plums (Valer et al 2007), cedar apple (Guillermin et al 2006), orange (Topuz et al 2005), pear (Gomez et al 2005), etc. The physical attribute such as size and shape, are considered in designing of drying and aeration systems, as these properties affect the airflow of the stored mass (Bern and Charity 1975; l az vi and Parvar 2007). Furthermore, daring the process of distribution and marketing, substantial Iosses are recorded which range frorri a slight loss of quality to total Spoilage. post-harvest losses may occur at any point in marketing process, from initial harvesting, grading, packaging, transportation from field to storage and storage to assembly point; during storage and distribution to final consumer. Causes of tosses are many such as physical damage during filing and trnnspnrt, ph riological decay, water loss, etc ((301. 2.013). In addition, postharvest diseases, which cause serious problems during storage and transportation of mango fruits, are the major factors that limit the thriving mango industry. A range of leaf, fruit, and soil disease can also affect mango fruits or cause to induce disease after harvesting. Some of the unacceptable defects/disease/disorder that cause the lowering of prices of fruits are anthracnose, latex stain/sap bum, stem-end-rot, chilling injury, fruit rot, cut wounds, stem end cavity, mealy-bug, hot water injury, mango scab, bacterial black spat, powdery mildew, etc. Barief descriptions of major diseases among these o feoncern are follows. 2.1.2. Posthrrvest disease of mangoes 2.1.2.1. Anthracnnse (black lesion) The major pre-and-pastharvset disease of mange is anthracnose caused by fungus Collcwfridiurn gfoeGsporioides (teleomorph: Gtomerdllo cin hffa) (Arauz, 2000). The fungus (C. gloeosporiode) that referred as anthracnose is responsible for many diseases on 11 not only mangoes but also on many tropical fruits including banana, avocado, papaya, coffee, passion fruit, and others (Nehson, 2008). According to Corl4di at aL (2006) anthracnose can result in serious decay of fruit during marketting and after sale, and may produce conspicuous, pinkish-orange spore masses under wet conditions. The temperature ranging from 20 to 3OC and a relative humidity > 95% favours the incedence of infections on panicle, lea' is and fruits (Dodd et al., 1991). In the field, anthracnose infections are favoured by wet, humid, warm weather conditions while postharvest anthracnose development generally favoured by warm and humid temperatures (Corkidi et al. 2006). The young fruit, when, infected in the field and infection remains in a quiescent stage until fruit ripening begins known as postharvest anthtracnnse (PruGsky, 1996) which causes rounded, black to brown lesions on the fruit surface. The fruit spots can and usually do coalesce and can eventually penetrate deep into the fruit, resulting in extensive fruit rotting. Fruits infected at mature stage carry the fungus into storage and cause cozi idorabie loss during storage, transit and marketing. However, sometimes in maturation phase, these are usually restricted to the peel and are generally evident during storage, Thus fruits that appear healthy at harvest can develop significant anthracnose symptoms rapidly upon ripening. SoieEimes However, most green fruit infections remain latent and largely invisible until ripening and ripen fruits affected by anthracnose develop only sunken, prominent, dark brown to blade decay spots before or after picking. But, in severe cases the fungus can invade the pulp and further advancement of disease causes development of acer uli and abundant orange to salmonpink masses of corlidia on lesions (Arauz, 2000). 2.1.2.2. Latex burry (staining) Sap burn, a serious quality cou m (Maqbeol and Maw, 2Ot 7), is the largest single quality problem with mango. Mango latex, a transparent, viscouse and clear or slightly milky fluid, is reponsible for allergic action on the skin Coveys et al. 1992) and can inflict serious eye effect (Menezes et al., 1995). Poor harvesting and handling techniques will result in a high incidence of sap burn which will severely downgrade the fruit quality (Sohsnsan and Parr, 2006). Sap is caustic in nature and has aroma characteristics of ripe fruit while oozes out instantly from unripe fruit as soon as the fruit is detached from staI1 . Sap can be separated into two phase Tower (milky and viscous) and upper phase (clear, yellow-brown 12 and oily) by centrifugation at 12000rpm for 5 min (Loveys et al,, 1992). Upper phase is more responsible for sap burn than lower phase. When sap comes in contact with the fruit during harvesting and postharvest handling causes undesirable skin blemish or bum (Anon, 2003). According to Bagshaw (1989), there are two type of sap: (1) spurt sap and (2) Ooze sap. Spurt sap released rapidly afar removing the stem and injuries the fruit skin within 5 second while ooze sap' released more slowly over about one hour but an=e sap does not injure the skin. In contrast, spurt sap would burs fruit, leaving a blemish that will develop during storage and transport (GO 1, 2013). Further, the amount and composition of sap exuded by a quit depends on fruit maturity, harvest time of day, cultivar (Bagshaw 1989), harvest season and production region (Rojas et al„ 1998). The less mature the fruit generally exuded more sap and the harvesting of fruit early in the morning has greater sap flow than later during the day (Bagshaw 1989). In order to avoid exudation of latex onto the fruit, the pedicel should be pointed downward at the time of removal (Anon, 2003), the fruit should be placed on de- sapping bench to drain out sap for 20-30 min and then fruit should be washed to avoid the chance of any sap burn (GOl, 2013). The maximum sap burn took place within the first 24 hour of sap contact but the injury continue even after 72 hour of sap contact in Chausa and `irdhuri cultivars of mango. T}u s extra care must be taken during first 24 hour of harvest. Dc-stemming under Ca(OH)2, Tweei -2A and Twcen-8Q can be most effective tie tmeits against sap bum in both cultivars (Maqbool and MaIik, 2007), 2.1.2.3. Stem end rat - Stem end rot, caused by several different fungi such as Dot iorella spp., Lasiodiplodia iheobrornae (Liao, 1975; Johnson and Cooke, 1991; Plaet et al, 1994) or PI wmwpsfs mangiferae (Ploetz et al., 1994; Ko et al., 2009), is also considered to the most severe post harvest disease of mango worldwide along with anthracnose. Several studies from several laboratories have been demonstrated that a complex of pathogens such as Bnrryosphneriicene (Slippers et al., 2005; de Oliveira Carta at al., 2010), including L. rheobrornae, N rnangjfrac, Neoflisicoccuin parvwn and Fusicoccuin aesculi, are associated with stem-end rot of mango- Far instance, F' aesculi (B.dothidea). N man iferae, N. parvwn causes stem-end rot of mango in Australia (Slippers et al., 005) and L. t cobrornue, F 13 aesculi. and N. pan'um pathogens are responsible of stem-end rat and dieback disease of mango in Brazil (De Oliveira Costa et al., 2010). Furthermore, A alrernaia, P. rrtang Brae and Botr}odiplodia spp. were reported by Amin et al. (2011) as the main pathogens that associated with stem-end rot of mango under the agro-ecologica% conditions of Punjab province of Pakistan. They identified fruit side rot as the major disease followed by stein- end rot among the p est diseases. However, this disease usually begins with the fungus attacking the stem of the fruit prior to harvest or through mechanically injured areas on the, stem or skin, where it remains quiescent until the fruit ripens. The initial symptom of stem end rat on harvested mature fruit is a darkening of the skin or circular black lesion around the base of the pedicel. The infected area may enlarge rapidly to form circular brownish-black areas of water soaked tissue which can extend over the whole fruit within several days. Once the fungus enters the stem end of the fruit, the whole fruit will rot within several days Anon, 2003). In addition at stem, the fungi can attack any part of the fruit especially that which becomes injured during harvesting or handling. It has been found that the Endophytic colonization of the inflorescence and pedicel tissue are the primary route of infection of fruit that develop stem-end rot during ripening. Storage at 3O°G favoured the infection by L. theobrorae over that by D. dom~nicanaas the cause of stem-end rot, and the reverse occurred at 25C and lower, which explains the possible different causes of the disease in tropical and in sub-tropical growing areas. Infection can be reduced leaving a pedicel of about 1-2 cm (Yahia 2005). 2.1.2.4. Chilling injury Chilling injury, physiological induced disorder (Yahia 2005), of mango fruit is occurring at storage temperatures below about 10-13° C (Mann and Singh, 1976) or 10° C (Hulsr , 1971) or i 1° (Anon, 2003) but according to II at al (1994), fruit stored at 8° C developed no chilling injury symploms. In contrast, Kane and Marcellin (1978) have reported the symptoms of chilling injury after 10 days storage of mango at 4 and 8 C -temperature and observed that the succinate oxidation capacity of mitochondria found to be decreased. Chilling injury, also, causes pitting and sunken lesions on the skin, uneven skin coloration, internal darkening of the pulp, off-flavor development, decay and leakage of the metabolites such as amino acids, mineral salts and sugars from the cell structure (Wills et al. 14 THESIS 1981). In another study by Medlicott (1990), chilling injury, as indicated by inhibition of ripening, was found at all harvest stored at 8°C, and in early season harvests stored at 1OC. Fruit from mid- and late-season harvests can be stored better at I 0C than at 12°C with no apparent signs of chilling injury. The severity of chilling injury, however, usually depends upon temperature and time of storage. For instance, storage at lower temperatures for longer durations causes more injury (Anon, 2003). Although, it has been reported that some cultivars-such or. Dashehari, Langra, etc. can be stored safely at 7-8CC for up to 25 days. The enhanced sesistan to the chilling injury might be related to the higher the total soluble solids content in fruit (Mukerhjee and Srivastava 1979). Further, temperature of the water used for hydra-cooling has been reported to be directly related to the appearance of surface chilling injury, so, it should be maintained at 10°C to avoid surface chilling injury ODeEll, et. al., 2000). The hot water treatment of Dasheri and L,angara fruits usually control the chilling injury during low temperature storage., Exogenous putrescine treatments can also be used to avoid the occurrence of chilling injury. Since, putrescine treatment inhibit ethylene production and maintained superoxide, disinutase activity in peril at a higher level and retard the malondialdehyde content and membrane permeability Zhang et al. 2000). In addition, storage of Aga fruits wrapped in microperfnrated poly ethylene or Xtend film in modified/controlled atmosphere of 5% CO2 and 10% 02 can also be reduced the possibility chilling injury (Pesis et al, 2002). Conditioning at high temperatures (35-38° ) for few lours bet'ore the low temperature can also be used for the better results (Yahia, 2005). - 2,1.2.5. Alternaria rot In the absent or well controlled of anthracrsose and stem-end rot disease, All aria rot which caused by A.lrernaria alternara can cause significant postharvest decay in mango (Pru.sky et aI, 2009). In other words, A. alternase, the casual organisms of black spot disease and stem-end rat, can cause serious/high Iosses during storage for three or more weeks (Yahia 2005) and compromise the storage Iife of fruits (Prusky at al. 1981, 1997; Kobiler et al. 1998, 2001), Through lenticels of fruits, A. i1ze17laze penetrates the fruit, darkens the intercellular spaces and collapses the cell. The appearance of disease consists of either small black spot (0.5-1.0mm in diameter) with dark centre and diffusive border or dark lenticels 15 (Yahia 2005). Initially, spots are concentrated around the stem end of the fruit where high numbers of leiiticels are present. The spots grow and coalesce to became a single spot that can cover half the fruit. Later, the disease progresses into the flesh which darkens and becomes partially soft (Anon 2003). Furthermore, Diedhiou et al. (2007) conducted a research to understand better the interactions between fungi involved in post harvest rotting of mango and local production practices and th6 changing climatic conditions during the maturation period. Their study showed that the Brits harvested during the humid season were more heavily infested and culturaI practices were payed an important role on mango infection whereby orchard sanitation and particularly cleaning and pruning reduced the infection rates. Orchards with no care, in contrast, yielded the most heavily infested mango samples. In addition, the harvest practice of inversion of fruits in soil for sap elimination increases contaniuiation with pathogenic fungi such as A1ternaria sp. B, tiienbromae, DothlarellQ sp., A. niger, etc. lansaur et al. (2006) isolated Afttrnaria alternata from rotted fruits of mango cuItivars Keitt, Kent, and Tommy Atkins alongwith Botdiplodia iheobrornae and Bovrytis c nerea - and proved to be highly pathogenic to all these varieties. In their stud', they treated fruits with HA (OC for 4 h) followed by 14 (40°C for 5 min) in combination and reported that the treatment was most effective for retarding postharvest disease without peel blackening and fruit damage and found to he increased the shelf life - of inoculated and uninocuiated fruits, The quality characteristics of non-inccuiated fl-its of the three varieties including total soluble, solids, titratable acidity and vitamin C contents were not significantly affected by these heat treatments. A11ernarii rot can also be controlled using combination of physical (15-20s spraying of hot water of 50-55°C temperature and brushing) and chemical (prochloraz, chlorine, etc.) treatments (Prusky et al. 1999). This new approach improves the fruit quality at the same as it reduced disease incidence (Prusky et al. 2009). 2.1.3. Packaging Packaging, a mean of convenient sized carriage, protects individual fruit from contact, rub and any type of compression damage. In addition, it avoids fruits from dirt, dust, pests, contaminants, etc. Wooden boxes, CFB (corrugate[) fibre boxes), polythene (low density polythene), crates, etc ire used for packaging of graded fruits (Gill et aL 2005). 16 Wooden boxes were commonly used for packaging and tniisxntation of mango fruits in domestic market but now due to the scarcity of wood and greater concern of the country over eiivironmentai core±±tion the use of CFB is the need ofhomr. The CFB boxes of 5 kg and 10 kg capacity for packing and shipping of mango fruits were designed and developed by the CISH (Central Institute for Subtropical Horticulture), Luknow successfully as an alternative to traditional nallcd wooden boxes (Medina and (Jarcia, 2002 ). The t sive use of CFB boxes has been r ported by Anon (2006) and Roy and Josbi (1989) fur export purposes. In addition, partitioned CFB boxes found to be best and more suitable for packaging and transporting of aiphonso cultivar (Roy and Pal, 1991) than wooden boxes. Paper scraps, newspapers, etc., ar commonly used as cushioning material for the - packaging of fruits which prevent them from getting bruised and spoiled during storage and transportation (Ravindra and Goswami, 2(J07). On the other hand, the low density polyethylene (LDPE) fining maintains humidity and results lesser fmit shrinkage during storage. Polyethylene packaging, thus, gave the best storage to the, products retaining the moisture content and sensorial attributes (Arnaud- inas, 1995). Similarly, the individual wrapping of fruits (Unpack) with newspaper or tissue paper an packing in honeycomb nets helps in getting optimum ripeiiing with reduced spoilage (Anon 2006, Narayana et al. 1991). But, sealing of mature green fruit fn semi-permeable polyethylene can cause quality deterioration in terns of taste and appearance (Straten and Osthuyse, 1994). Thus, the use of clean, new and qualitative materials, particularly of paper or stamps bearing trade specifications should be allowed and provide the printing or labeling with non-toxic ink or glue so that any type of internal or external damage can be avoided. Mangoes shall be packed in each container in compliance with the Recommended International Code of Practice for Packaging and Transport of Fresh Fruits and Vegetables (CAC, 2004), 2.1.4_ Storage Storage is essential for extending the consumption period of fruits, regulating their supply to the market and also for transportation to Lang distances. The main objective of storage is to arrest damage and extend the shelf life of fruits. Various storage techniques such as love temperature storage, low pressure storage, CA/MA (controlled 1] atmospherelmodifed storage) and use of chemicals, wraps and coatings and inniding radiation for horticultural produces have been emerged so far. The mature green fruits can be kept at room temperature for about.-4--10 days depending upon the variety For exports, the harvested fruits are pre-cooled to 10-12CC and then stored at an appropriate temperature. The fruits of Dashehari, Mallikaka and Amrapali should be stored at 12C, Langra at 14C and Chausa at 8°C with 85-90% relative humidity. The fruits could be stored for 3-4 weeks in good condition at low temperature. According to Kader (2002), temperature between 13°C for mature green mango and 10°C for partially ripe and ripe mangoes is generally recommended with 90-95% RH. The 90% RH and 10-15° C t nperature from ambient condition cnuld be achieved in zero enemy coal chambers (ZECC) by watering the chamber twice a day. The ZECC was developed by IAtL New Delhi, -using locally available materials such as bricks, sand, bamboo, dry grass, jute, etc. works on the principle of evaporation. After performance evaluation of ZECC in various parts of country, it has been reported that the shelf Life of mango is increased by 3-4 days (Roy and Pal, 1991). Economical, in terms of construction and operation, and its satisfactory performance resulted developments of a variety of various ZECC models of capacity between 100 kg and 1000 kg. These were tested and found performed equally well in reducing postharvest losses and maintaining quality of fruits and ' egetables. Improved type ZECC is one of them (Kitinoja, 2010) and satisfactory storage at 7.8°/(s °C has also been reported (Mann and Singh 1976). In spite of, for better results, CA/MA (controlled atmosphere/modified atmosphere) can be recommend d. Since, CA provides an effective storage environment (Sender et al. 2000, Raghavan et al 2003) and MA can be used to maintain postharvest quality of fruits (Ding at al 2002, Rndov et al 2002). Controlled atmosphere (3-4 % CO2and 4-5% 02) storage under a continuous flow system held at 13-15°C for Alphonso mango in which fruit could be kept for 0 days with a post storage ripening period of 4 to 5 days. Lalel et al, (2001) has recommended, 6% CO2 and 2% 0,, atmospheric composition of CA for extending the shelf life of mangoes (cv. Kesin ton Pride). For CA/MA storage of mango, an atmospheric composition of 4-10%, CO2 and 1-8% 02 have been studied and recommended by Lalel et al. (2003, 2005), Bender et al. (7-000), 18 Sh'ukor et al, (2000), etc. in varying combinations for a givers cultivar and maturity of fruits. For instant, optimum matured fruit dv no exhibit symptoms of chilling injury, have longer storage shelf life and may withstand maximum limit (25%) of CO2 (Bender et al. 2000) showed insecticidal effects during storage (Yahia Vazquez-Merino 1993). However, storage of mango in the atmospheric composition of 15% 0O1 and 1% 02 results quality degradation in terms of off-flavor and skin discoloration. Modified packaging inhibited the mango ripening process (ornsrivichai et al. 19 2) . 2.1.5. Transportation Mango transportation needs careful handling to avoid the fruits from damage and also spoilage. In addition, during transportation fruits is ripening become too soft, easy injury that reduces eating and keeping quality of fruits. Thus careful inspection of the empty truck or vans is needed to ensure that these are free from pests and plant debris prior to loading with treated packages of mango. For this purpose, a joint inspection by the Directorate of Plant Protection, Quarantine &' Storage, Ministry of Agriculture, NH IV, Faridabad and Animal & Plant Health Inspection Service US Department of Agriculture is carried out before exporting the fresh fruits of mango from India to USA. The second disinfection of the empty van or tuck is required if any pests are found in the first round inspection. _Using suitable insecticide, empty van or truck thoroughly disinfected to ensure that the pests are effectively controlled, The space between the doors of van and loading area of the facility will be covered by insect-proof screen to prevent entry of hitchhiking pests during koading and after completion of loading the doors of the van or truck is closed and secured by a Lack and a seal will be affixed (Anon 2007), 19 2,2. Principle and Potential of non-destructive food quality evaluation techniques The quality evaluation in the food industries is still heavily depends on manual inspection, which is tedious, laborious, and costly, and easily inf]ttenced by physiological factors, inducing subjective and inconsistent evaluation results (Du and Sun, 2006a, b). So. to achieve a valuable quality control for food products, it should not be dependent an human errors dt,ring iispeeton, and therefor, quatity control systems must be automated. For instance, ' differences between computer vision (Fig, .1a) and human vision system (Fig.2.1b) have been displayed in Table 2.33. Since, if quality evaluation is achieved automatically, production speed and efFicienc ► can also be improved. In addition, evaluation accuracy can be increased accompanying with reduction in production cost (Sun and Brosnan, 2003a). Consequently, computer vision systems have been used increasingly in many food industries, such as sugar, wheat and vegetables industries (Marquez and Anon 1986 McDonald and Chen, 1991) for quality evaluation purposes so for as a rapid, economic, consistent and even more accurate and objective inspection tool (Sun 2008, 2000). Although, both oiiter and inner quality information can be collected by an automatic grading system in a factory, but computer vision is more effective for measuring outer parameters (Lee et al. 1999; Majumdar and Jayas 20(710a, b; Shahin and Symons 2001; Paliwal et al. 2003; Shigeta et aI. 2004; Shahin et al. 2004; Qiao et al. 2004; Lorestani et aL 006). A computer vision system generally comprises five basic components: illumination, a camera, an image capture board (frame grabber or digitizer), computer hardware and software (Wang and Sun, 2002). Software is recognized as being core of computer vision system among them. Since, with the help- of software, the two very important tasks in computer vision: image processing and image analysis, are carried out. Generally, image analysis extracts information from practised images and consider as high level image processing. Image processing techniques based on computer vision system are being used increasingly in the field of agricultural and food products for quality assurance purpose. lor 20 instal ce, an automated quality verification system (Nf oroge et al., 2002) and a multiproduct grading system (I .undo et al. 2005) were proposed for agricultural products. In addition, to- Camera Fig,2.1a. Diagram of typi computer visioi (Abdullah 2OO C Fig.2.1 b. Human vision system A -estimate-size, sort-colour, classify-shape, detect-braises or scar-tissue and to predict-the mass of the fruits, a number of algorithms have been developed and applied so far during image processing. Image processing modifies pictures to improve them (enhancement, restoration), extract information (analysis, recognition), and change their structure (composition, image editing. Images can be praces~ed by optical, photographic, and electronic moans, but image processing using digital computers is the most common method because digital methods are fast, flexible, and precise (Anon 1996). Nowadays, the use of image processing is gaining interest for the evaluation of internal (sweetness, acidity or inner diseases), external (size, color intensity, color homogeneity, bruises, shape, stem identification, surface texture and mass) and freshness quality of agricultural produce (Table 2.4). 21 Table 23: Machine vision versus human vision: capabilities and performance evaluation Capability Machine vision system Human vision system Capabilities of evaluation - Distance Limited capability Good quality capability Orientation Good for 2-D Good quality capability Motion Limited, sensitive to image blurring Good quality capability Edgeslregions High contrast image required Highly developed Image Special software needed: limited Highly developed organization Capability Surface shading Limited capability with gray scale Highly developed scale 2D interpretation Excellent for well defined features Highly developed 3D interpretation Very limited capabilities Highly developed Overall Best for quantitative measurement Best for qualitative structured scene interpretation of complex, unstructured scene Performance of evaluation Resolution - Limited by pixel array size High resolution capability Processing speed Fraction of second per image Real-time processing Discrimination Limited to high contrast images Very sensitive discrimination Accuracy Accurate for discriminating quantitative Accurate at distinguishing differences, accuracy remains consistent qualitative differences; may at high production volumes decrease at high volumes Operating cost iligli for low volume; lower than Human Lower than machine at low Vision at high volume volume Overall Best at high production volume Best at low or moderate volume Thus, the application potential of image processing techniques, to evaluate food quality iii food industries, has long been recognized. 22 Table 2.4 Different components of qualities for fruits and vegetables Compoacnts QuallLlcs External size Weight, volume and dimension Quality Shape Diameter/dcpth ratio Colour Uniformity, intensity Defect Bruise, stab, spot Flavour Sweetness, Sourness, Are=a, Astringency Internal Texture Firmness, Juiciness, Crispness Qualities Nutrition Carbohydrates, Proteins, Vitamins, Functional properties Defects Internal cavity, water core, frost damage, rotten Now, non-destructive techniques for quality evaluation have gained momentum (Iwamoto et al., 1995; Jha and matsuola, 2000): Since, these techniques, particularly for fruits and vegetables, are quick and easy to use (Jha et al. 2004). Many physical characteristics of fruits and vegetables have been determined non-destructively ( ato, 1997) in last decade. Due to the fact that most p arts of the external quality attributes are currently inspected visually, machine vision provides a means to perform this task automatically (Molto' et aL, 1991; Diaz et aI_, 2000). New technologies, such as new Hardware architectures, are making possible the implementation of these kinds of application in real- time (Crowe and Detwiche, 1996a, b; Aleixos et aL, 1999. With a global supply of food on the marketplace, different non-destructive testing methodologies have been emerged and some of them are: computer vision, X-ray and computed tomography, magnetic resonance .imaging, near infrared, ultrasound and electronic nose. Of these, some are used on different food items while ethers are specific to a particular fare. 2.2.1. X-ray and computed tomography Electromagnetic waves with wavelengths ranging from Ito I00 um are called soft -rays, - rays have not yet been used to their fullest potential in a range of application areas, especially the agricultural and food industries. However, X-ray techniques are gaining momentum and becoming an alternative to other imaging techniques because of tow penetration power and ability to reveal the internal density changes that makes soft X-rays 23 suitable to be used for agricultural products. The soft X-ray method is rapid and takes only a few seconds (3-5 s) to produce an X-ray image. When an X-ray beam passes through the matter (sample), it undergoes attenuation. Due to attenuation, the intensity of the X-ray energy decreases gradually by absorption and scattering. Absorption refers to the case in which an incident X-ray photon gives up all of its energy. Scattering refers to those X-ray photons that have undergone a change in direction after interaction with atans of matter. Other photons which are neither absorbed nor scattered, simply pass through the matter and can be detected. This process is known as transmission X-ray imtging. So, X-ray imaging is a transmission based technique in which X-rays from a source pass through the food products and are detected either by film or an ionization chamber on the opposite side of the product. SimilarIy, computed tomography or CT imaging/scan is an x-ray procedure that combines many x-ray images with the aid of a computer to generate cross-sectional views and, if needed, three-dimensional images of the internal organs and structures of the product's body. CT is the imaging method that can produce the highest resolution angiographic images. So, the X-ray computed tomography (CT) is a technique that uses X- ray► images to reconstruct the internal micros eture of objects and to monitor the internal quality changes in foods. For instance, the physiological constituents have been monitored in peaches by CT methods in which x-ray abscabed by the peaches is expressed in CT number and used as an index for measuring the chap s in internal quality of the fruit. Relationships between the CT number and the physiological contents were determined and it was concluded that x-ray CT imaging could be an effective too] in the evaluation of peach internal quality, 2.2.2. Magneticresonance imaging (f4 Magnetic resonance imaging (MBI) previously known as nuclear resonance imaging NMR), gives the density of protons or hydrogen nuclei of the body at resonant frequency. [Jnlik.e CT. MRI provides excellent renditions of soft and delicate materials. This unique characteristic makes Mlkl suitable for visualization of most food objects, and applications range from non-invasive to real-time monitoring of dynarrdc changes as foods are processed, stared, packaged, and distributed. 24 The principle of MRI is based on the association of each spatial region in a sample with a characteristic nuclear magnetic resonance frequency, by imposing an external magnetic field. The purpose external magnetic field is to get net magnetization. Since, without the external magnetic field, the magnetic mom~esnt would point i-x all directions at random, and there would be no not magnetization. But, in the presence of a large gnelic field, the hydrogen nuclei will preferentially align their spin in the direction of the magnetic field which is known as the Lamar effect, and the frequency at which the nucleus proceeds around the axis is termed the Lamer frequency. This effect implies a transfer of energy from the spin system to another system or lattice. The transfer of energy is chamcteied by an exponential relaxation lawn with time constants Ti and T2 which are also known as the spin— lattice excitation and spin—spin relaxation times, respectively. In commercial . RI, the magnetic field ranges from 0.5 to .4 (compared with Earths magnetic field of less than 60pT) tesla. Ti is typically of the order of 0.2-2 s, and 72 ranges from 10 to 100 ms. Fig. 2.2 shows the basic components of MRI system, Fig. 2.2. Block diagram of an MRI system (Jiti 1989), Magnet Graaaent Coif GrdIenI Amp FF Ca7 canao+ sample Neural RF AmP RFaB~ RF Recei~ren DIGNber Car ;I COi[ Magnet MRI system comprises a scanner, a static magnetic field and RF coils which are used to transmit radio-frequency excitation into the material to be imaged. This excites a c nponent of magoeti atiorl in the transverse ply which cail be detected 'by a RF reception coil. Thus, by applying a radio-frequency ( field at the resonant frequency, the magnetic moments of the spinning trncloi lose equilibrium and hence radiate a signal which is a function of the line integral of the magnetic resonance signature in the object. This radiation reflects the distribution of frequencies, and a Fourier rsforna of these signals provides an image of the spatial distribution of the magnetization (Sun 2009). 25 Owing to the fact that MEl provides rapid, direct, and., most importantly, aaninvasive, non-destructive means for the determination of not only the quantity of water present but also the structure dynamic characteristics of the water, this r lively new imaging technique has become useful for food engineering. There are numerous applications of MRI in agriculture, since water is the basic building black of many food materials. The simplest among them is the determination of moisture and oil content, In addition, the ice formation during food freezing can be examined using MRL methods as the formation of ice has been seen to reduce the specially located NMR signal. The characteristics of a food can be better controlled as MRI can serve to assess freezing times and the food structure during the freezing process. Other interesting applications include real-time monitoring of ice gradients in a dough stick during the freezing and thawing processes, mapping the temperature distribution patterns in food sauce during microwave-Induced heating, and predicting sensory attributes related to the texture of cooked potatoes. 2.2.3. Infrared The electromagnetic infrared light radiation range lies in the region of 600-IOI)Onm and the technique responsible for generating images with infrared light known as thertnographic photography. The principle of thermographic imaging is based on the fact that all objects emit a certain amount of thermal radiation as a function of their temperature and the object with higher temperature emits the more IR radiation. This radiation is detected by a specially built camera known as an IR camera that is very similar to the ordinary camera for visible light. There are several applications of the IR imaging in the food industry -- such as identification of foreign bodies in food products and may major physiological -properties of foodstuffs (firmness, soluble-solid content, and acidity) appear to be highly correlated with IR signals, implying that image analysis of IR thermography is suitable for quality evaluation and shelf-life determination of a number of fruit and vegetable products. Thus, thenmI imaging offers a potential alternative technology for non-destructive and non- contact image-sensing applications. Good thermagraphic images can be obtained by leaving the object at rest below the IR camera, applying a heat pulse produced by a flashlight, and monitoring the decreasing temperature as a function of time. Because of different thermal capacities or heat conductivities, the objects will cool down at different speeds; therefore, the 26 thermal conductivity of an object can be measured by the decreasing temperature calculated from a sequence of IR images. Pkbchfruii LuJcwtqaitiit PC - 5k eM fra inSmftw,ne Opdci1 FLPrqllt huiille a pIcr~ Fig. 2.3, Schematic diagram of the experimental setup for NIR tasting of peach fruit (Capone et al. 2005) Near-infraired spectroscopy method is no longer new; as it started in early 1970 in Japan just after some reports from America. Thi.s method use as rapid and often nondestructive technique for measuring the composition of biological materials has been demonstrated for many commodities, In Japan, NIR. as a nondestructive method was started for the determination of sugar content in intact peaches, Satsuma orange and similar other soluble solids. To determine the solid content of cantaloupe NIIR light at 884 nm and 913nm were used. Initially the correlation of their findings was poor mainly due to light lasses. Later, it was modifies and applied it to honey dew melons; the improved methods showed better correlation. Similarly, a nondestructive optical method for determining the internal quality of intact peaches and nectarines was investigated. A schernatzk diagram of the experimental setup for NIR testing of peach fruit is given in Fig. .3. Based upon visible and near-infrared spectrophotometer techniques, the method was capable of simultaneously predicting the soluble solid cnte.nt, sucrose content, sorbitol content, etc. of intact peaches and nectarines and that too without much sample preparation. Some modifications in spectrophotometers, especially for holding the intact sample such as liquid food, are reported. Also, recently a low cost I4IR spectrometer has been used to estimate the soluble and dry matter content of kiwifruit, acid-brix ratio of tomato juice (Jha and Matsuoka, 2004). 27 2.2.4. Ult asoun i Changes in acoustic properties that can kbe related to density changes in food product is the main reason of growing interest to use ultra sound , for food quality evaluation (Mc lements, 1995). In addition to this, ultrasound has the ability to differentiate between the propagation velocity within various media, and the differences in acoustic impedance between different regions within a given volume.. Tho information, ultrasound provides, generally originates from the reflection or transmission of sound waves emitted - by an external source. Refraction, absorption, and scattering also play a role, but mainly as factors that degrade the ultrasonic measurements. A typical sourcedetector unit is based on a piezoelectric crystal resonating between 1 MHz and 10 MHz. The basic physical parameters of importance are the frequency of the wave and the acoustical impedance of the object through which the sound wave travels. In the past, ultrasound has been used for measuring the moisture content of food products, predicting the intramuscular fat content of bovine products and studying and evaluating the turgidity and hydration of orange peel. In addition to these, one of the most widespread and promising ultrasonic applications is for composition measurement. Recent studies have shown that ultrasonic velocity measurements can accurately be used to predict the fat, water, protein, and other chemical compositions of meat-based products. In spite of several advantages, there are three co on drawback of this technique as: (1) low image spatial resolution — typically of a feu millimeters, (2) low signal-to-noise ratio and (3) many artefacts. Despite these drawbacks, the technique is safe and relatively inexpensive. Current research methods tend to elinninate artefacts, improve image contrast, and simplify the presentation of data, and many efforts are being directed towards three-dimensional data acquisition and image representation (Sun, 2008), 2.2.5. Electronic nose (EN) The basic principle of EN system (Fig. .4) is totally based on sensors that create a unique smell print and respond to the whole set of volatiles in a unique digital pattern after considering the total headspaee volatiles. 'These patterns are signature of particular set of aromatic compounds. For each process or application of interest, a database of such digitized patterns is created, called the training set. When an unknown sample is exposed to EN sensors, the EN first digitizes the samples volatiles and than compares it with the existing 28 training set. Recently, several applications of EN have been investigated such as: maturity indicator for apple and tea taster. Commercially available ENs use an array of sensors combined with pattern recognition software, There have been several reports on electronic sensing in environmental control, medical diagnostics and the food industry. Some authors reported positive applications of EN technology for discriminating different fruit's quality and many experiments were performed including testing orange, melons, blueberries, pears, peaches, bananas, apples and nectarines. it can be concluded that the RN, for nondestructive quality evaluation of foods, has great potential especially for fruits. Fig, 2.4: Image of the portable Electronic Nose developed at IMM-CNR — Lecee, Italy (Capone at al. 2003. In making physical assessments of agricultural materials and foodstuffs, images are undoubtedly the preferred method in representing concepts to the human brain. Many of the quality factors affecting foodstuffs can be determined by visual inspection and image alysis. Such inspections determine market price and, to some extent; the "best-used- :fore" date. Machine vision systems are ideally suited for routine inspection and quality surance tasks, Backed by powerfat artificial intelligence systems and state-of-the-art .ctronic technologies, machine vision provides a mechanism in which the human thinking ocess is simulated artificially. To date, machine vision has extensively been applied to lye various food engineering problems, ranging from simple quality evaluation of food oducts to complicated robot guidance applications. 2.2.6. Computer vision system Machine'computer vision is a relatively young discipline with its origin traced back to the 1960s (Waxes 1994). Following an explosion of interest during the 197Qs, it has experienced continued growth both in theory and application. Sonka et al. (1999) reported 29 that more than 1000 papsrs are published each year in the expanding fields of computer vision and image processing. Machine vision is an engineering technology that combines mechanics, optical instrumentation, electromagnetic sensing, digital video and .image processing technology. As an integrated mechanical-opt cal-electronic-software system, machine vision has been widely used for examining, monitoring, and controlling a very broad range of applications. It is the construction of explicit and meaningful descriptions Qf physical objects from images (Ballard and Brown 1952) and it encloses the capturing, processing and analysis of two-dimensiornal image çnmmermaiis 1998). However, in another study by Sonks et at (1999) noted that it aims to duplicate the effect of human vision by electronically perceiving and understanding an image, arid provides suitably rapid, economic, eon.istent and objective assessment (Sun 2000). So it can be say-that the m'chirte vision is the use of devvices for optical, non-contact sensing to automatically receive and intermit the image of a teal scene in order to obtain information and/or control machines or processing of image (Zuech et al. 2000). The food industry continues to be among the fast growing segments of machine vision systems (Gunasekaran 1996). It has also been used successfully in the analysis of grain characteristics and in the evaluation of foods such as potato chips, meats, cheese and pig. Crowe and Delwiche (1996a, 1996b) have developed a machine vision system for sorting and grading of fruits based on color and surface defects. Tao et al. (1991) and Tao (i 996b) have developed a high capacity color vision sorter based on optical properties such as reflectance to determine colour, shape, size, textural feature (Mlajumdar et al. 2000a, 2000b, 0ODc, 2000d), volume (Wang and Nguang 2007) and surface area ifert et al. 20(5; Thaler et aL 2007) of fruits and vegetables including apples, peaches, tomatoes and citrus. Computer vision systems provide suitably rapid, economic, consistent and objective assessment; they have been used increasingly in the food and agricultural industry for inspection and evaluation purposes (Sun 2000). They have proved to be successful computer vision system for the objective measurement and assessment of several agricultural products (Ticnmermaus 1998). Over the past decades, advances in hardware and software for digital image processing and has been motivated several studies on the development of these Systems to evaluate the quality of diverse and processed foods (Locht et al. 1997; Gerrard et al. 1996). In addition, it has long been recognized as a potential technique for the guidance 30 or contra' of various agricultural and food processes (Tillett 1990). The extensive studies, thus, have been carried out over the past 20 years artd consequeutly generating manly publications. Nevertheless, we can say that the computer vision technology not only provides a high level of flexibility and repeatability at a relatively low cost, but also, and more importantly, it permits fairly high plant throughput without connpromising accuracy. Applications Pf these techniques have new been expanded to various areas .such as medical diagnostic, automatic manufactl1ring and surveillance, remote sensing, technical diagncstios, autonomous vehicle, robot guidance and in the agricultural and food industry, including the inspection of quality and grading of fni t and vegetable. Some impottant applications of machine vision have been given in the Table 2.5 (Singh et al. 2004). The majority of these studies focused on the application of computer vision to product quality inspection and gradirrg. Traditionally, quality inspection of agricultural and food products has been performed by human graders. However, in most cases these manual inspections arc time-consuming and labour-intensive: Moreover the accuracy of the tests cannot be guaranteed (Park et al. 1996). By contrast it lies been found that computer vision inspection of food products was more consistent, efficient and cost effective (Lu et al. 2000; Tao et al. 1995a). Also with the advantages of superior speed and accuracy, computer vision has attracted a significant amount of research aimed at replacing human inspection. Recent research has highlighted the possible application of vision systems in other areas of ag~ieulture, including the analysis of animal behaviour (Sergeant et al, 1998), applications in the implementation of precision fnrniuig and machine guidance (Tillett and Hague 1999), forestry (Krum et a]. 2000) and plant feature measurement and growth analysis (Warren 1997). Besides the progress in research, these is inereasitg evidence of computer vision systems being adopted at commercial level. This is indicated by the sales of ASME (Application Specific Machine Vision) systems into the North Ameri n food marker, which reached 65 million dollars in 199 {Lacht et a.. 1997). 31 Table 2.3: Summary of machine/computer vision applications for different areas Application areas Purpose Industrial automation Process control, quality cor tool, geometrical measurement and image processing Barcude and package label reading, object sorting Parts identification on assembly lines, defect and fault inspection Inspection of printed circuit boards and integrated circuits Medical image Tumor detection, measurement of size and shape of internal organs, analysis blood cell count X-ray inspection Robotics Obstacle avoidance by recognition and interpretation of objects in a scene Collision avoidance, machining monitoring Hazajd determination Radar imaging Target detection and identification, guidance of helicopters and aircrafts in landing, guidance of remote piloted vehicles (RPV), guiding missiles and satellites from visual cues Food industry Sorting of vegetables and fruits, location of defects e.g. Iocation of dark contaminants and nsects in cereals Document analysis Handwritten character recognition, layout recognition, graphics recognition Singh at al. (2004) Graves and Batchelor (2003) summarized more than 20 machine vision applications that were classified by tasks in the natural product industry, more than 15 in manufacturing industry, and 7 other machine vision tasks applied to various situations such as security and surveillance, medicine and health screening, military, and traffic control and monitoring. Gunasekarait (1996) reported that the food industry is now ranked among the top ten industries using machine vision technology. This paper reviews the latest development of computer vision technology with respect to quality inspection in the agricultural and food industry. 2.3. Machinelcompufer vision system and its principle components A computer vision system generally consists of five basic components: illumination, a camera., an image capture board (frame grabber or digitizer), computer hardware and software (Sun 2000; Wang & Sun 2001, 2002a) as shown in Fig. .5a. However, the mechanical design for a specified machine vision system usually is uniquely structured to suit the inspection of a particular pPro h ct, For instance, the Conveyor belts in a poultry bone detection machine vision systems are usually flat; with no texture, and made of IJSDA approved plastic materials. While prototype machines for, chicken defect and disease inspection in the chicken plants use hooks to hold the birds when they are dangling from moving chains and passing through light beams (Chao et al. 2000). On the other hand, most conveyor belts for -a modem apple packing line are roller conveyors made of special dark colored rubber (Tao 1996). Thu lighting system, a critical part of a conlralled machine vision systen, rnuSL be carefully designed. The ultimate purpose of lighting design is to provide a consistent scene eliminate the appearance of iariaXions, and yield appropriate, application-spec.i f e lighting. Proper selection of lighting sources (incandescent, fluorescent, halogen, Xenon, LED), lighting arrangements (backlighting, front lighting, side lighting, structured lighting, ring Iighting), and lighting geometry (point lighting, diffuse lighting, collimated lighting) is the "key to value" (Zuech 2004). Primary factors that influence the selection is whether the object under inspection is: 1) flat or curved; 2) absorbing, transmissive or reflective; and 3) the nature of the feature to be imaged in m rison with the bacicgrnun& For instance, backlighting is usually used for detecting objects that are translucent, such as hatching eggs (Dos and Evans 1992) or used to measure the geometric dimensions of obscured object, such as measuring the shoot height or root diameters of pine to estimate, the pine seedlings (Wilhoit et al. 1994). The side light scatter can be used to determine the cellular granularity (fain et al. 1991). Structured light can be used to farm a 3-D shape of apple surfaces for stem/calyx detection (Yang 1993). Moreover, controlled lighting design sometimes acts as an active sensing means that can "create" new information. Laser stripes of structured lights combined with X-ray imaging on deboned chicken meat inspection is a successful example of generating "extra" information in machine vision systems. Because of a controlled lighting system design, intelligent image processing technology appIied to a machine vision 33 system is normally simplified and can achieve high accuracy. Those algorithms tend to maximize the utilization of pre-obtained object properties such as the appearance, geometry, surface issues, shape, size, color, and positions, as well as the effect of lighting sources, For instance, the rule-based decision-mating method, which is quite effective under controlled condition (Bartlett 1988), is rarely used in the computer vision pattern recognition field due to uncontrolled possibilities. In short, a major key to a successful machine vision application is to start with a good contrast, repeatable image that is not affected by ambient light or the surroundings. A functional block diagram of basic machine vision system has been given in. Fig. 2.5 [b). a y me UW PC Fig 2.5 (a) Principle components of machine vision system (Punditet al. 2005) Fig 2.5 (b) Functional block diagram ofbasic machine vision system FM (Singh et al. 2004) M Image „D _. Mare Senior I lm2ge U Cane 1(r Oprrata, UeCi inn Qinsale Makie~ Display 34 2.3.1. Image acquisition system Image acquisition or capturing is an application program enables users to upload pictures from digital cameras or scanners which are either connected directly to the computer or to the network. For the grading, classification and analysis of most agricultural images, only 2-dimensional (D) data are generally needed. However, for more information on structure or added details in many applications using image analysis technique, the 3-D image is required. For instant, S 2', .I. Ultra violet imaging system Ultraviolet (UV) radiation, the portion of electromagnetic spectrum between x rays and visible light (or between 40 and 400 nm i.e. 30-3 eV), is generally divided into Vacuum UV (40-190 am), Far UV (190-220 nm), LI (220-290 nm), LT B (290-320), and UVA (320-400 urn). In spite of sun as primary and natural source of UV radiation, there are some artificial sources of UV nidiatioa such as tanning booths, black lights, curing lamps, germicidal Iamps, mercury vapor lamps, halogen lights, high-intensity discharge lamps, fluorescent and incandescent sources, and some types of lasers, also. The use of these artificial resources generally depends on the requirement of the wavelength responsible for hazardness and which makes basis for applications (Anon 2009). The wavelength of the UV light is not shifted during the application. Since, ultraviolet light interacts with materials in an unique way, enabling features and characteristics to be observed that are difficult to detect by other methods. Reflectecl-UV imaging starts with the fflumirration of a surface with ultraviolet light and reflection or scattering of UV light and then image acquired by the camera sensitive in UV band. In this process, the UV light tends to be absorbed strongly by many materials and makes possible to visualize the surface topology of an object without the light penetrating into the interior parts. However, surface features that are not apparent at longer wavelengths, can be scattered by UV light because of short wavelength. Similarly, UV-fluorescence imaging also starts with active illumination with IJY light, but the detected signal is in the visible or infrared (IR) baA The fluorescent material absorbs the U excitation then reradiates at a longer wavelength. The emitted fluorescence is not reflected light, it tends to be a diffuse emission (Richards, 2006). Several studies have shown that the LTV light (365nrn) imaging system can be used for the purpose identification of fruits and vegetables with moderate to severe leveIs of freeze damage. For instance, Slaughter et al. (2O)8') found small dot pattern visible when the .peel of oranges illuminated by long wave of UV light (365 nm). Hagen et al. (2006) tested chlorophyll fluorescence (h1F) as a tool for non-destructive estimation of the flavor-old content in the skin of apples. Dugo et al. (2005) found a pattern of 1-201m bright yellow dots on the peel of freeze-damaged oranges, grapefruits and tangerines by using long wave ISV light (365 nm) and reported that the fluorescent pattern is likely due to the fluorescence of tangeritin, a poly-methoxylated flavones in the peel oil of citrus fruits. Katsumata et al. (2004) studied on visible photoluminescence polished rice, flour, peanuts, barley astd other starches and reported that the application UV light acts as safe, efl`ective, and powerful non destructive tools for quality control of the foods. In addition, atsumata et al. (2006) also had developed a non-destructive technique for rice based on pbotolurninescence and they examined it with various temp Lures. According to them, the peak intensity was found extremely sensitive towards the temperature of spevin ens and peak intensity of PL decreases exponentially with temperature. Similarly, Persson (1998) studied on image analysis of shape and size of fine aggregates using UV light and reported that there is a major and 36 distinct difference between the shape of crushed and natural aggregates. UV light based image analysis of aggregates, thus, gives multiple possibilities of churacethition and classification Visual inspection of citrus under UV illumination (i.e. in `black light" rooms), thus, could be used to remove fruits subjected with fungal infection and the yellow spot pattern' associated with the LTV fluorescence of freeze-damaged citrus peel tissue have a unique and high contrast color compared to undamaged peel tip that allowed simple and high-speed image processing techniques could be used in quantifying the level of damage (Slaughter et al. 2008). Another potential advantage of the UV fluorescence technique is, it could be implemented fairly and quickly by human inspectors in existing black light inspection rooms in parking lines, requiring only a small amount of additional training and no additional capital expenditures. Currently, visual inspection is done in black light rooms to remove fruit with fungal infection where training inspectors can identify fruits with small yellow fluorescent spots that should be feasible. So, this method has been found suitable for both visual inspection using black light inspection rooms as well as for automation using machine vision techniques by the researchers. 2.3.1,2. VilbIo imaging system Richards (2006) illustrated that the industrial machine vision system, which prime objective is to replace human (whose visibility lie between 400 and 180 run in the portion of the electromagnetic spectrum) work, was traditionally centered on visible-light imaging cameras and visible-light illumination. The most important application of visible imaging_ system is in qualitative sorting to minimize losses of fruits and vegetables by judging physiological disorders and surface blemishes. Several studied have been reported by the researchers on the application of visible imaging system. For instance, Knight et al. (2002) explained physiological peel disorder in citnis caused by mechanical peel damage and reported that the frame damage disorder olcocellosis (rind-oil spot), due. to d pression of tissue surrounding the individual glands, is a surfbce blemishes, found visible when illuminated by visible Tight. On the basis of their visual characteristics and significant correlation between peel color score (visual assessment) and Hunterl ab alb ratio can be used for classification of oranges and for maturity estimation of mangoes (Med1ieott, at al. 1992). 31 Another approach of visible imaging is the monochrome imaging (i.e. image analysis of black and white image). Monochrome camera often can provide a greater range of digital information than color images (Mnrceau and Francois 2009). For analyzing monochrome images, digital color images transformed into grayscale, are considered superior. Yang and Tillett (1994) has also presented monochrome approach and reported that the monochrome digital images offer a greater range of information than color images due to the filtering tethniques used to acquire color images through digital formats. In their study they concluded that the intensity of fruits decreases from centre to boundaries due to fruit curvature and the defects usually appear darker than the rest of the parts of fruit. In addition, the contrast between healthy and sick tissues also changes from one defect to another in monochrome imaging. In spite of several advwitagos of monochrome imaging, the digital colour images were also-found superior for some image analysis applications due to the fact that they are hue to life and have capability to discriminate and detect russet, defects and contaminations. For instance, Heinemann et al. (1995) discriminated russet by colour camera using the hue (H), and a hyper spectral imaging technique for the detection of apple surface defects and contaminations developed by l chi et al. (2004). Visual spectral analysis was used by Tha et al. (2005) to analyze the physical characteristics of mangoes and Harrel (1991) generated a Bayesian discriminant model for sorting fruit and vegetables. For grading of mangoes, the sweetness of intact mango was determined using visual spectral analysis by Teoh et al. (2006). 2.3.13. NIR Imaging system Near-infrared (N R) imaging, a fast and nun-destructive analytical technique, provides chemical and physical information of any matrix. NI imaging system combines the capability of detecting not only the presence of chemical species but also pinpointing of their location. In place of a single specirarn, N[R imaging siiiItaneously generates tens thousands spectra, each relating to a specific area of the sample. Pb*erful statistical analysis tools extract .pertinent information from the resulting datuset. NIR`imaging systems can collect more than 80,000 spectra in minutes, with user-friendly processing software and raw data into usable information. Image generation allows a qualitative- overview of a sample, which can be useful for making rapid comparisons 38 Delwiche (1987) has developed a hardware to acquire color and near-infrared (NIR) images of a peach moving on a bait anveyor and r orted that the NIR images were capable of detecting the defects only while color images were used for finding of maturity, amount of blush, and some defects in the grading of peach. Similarly, Miller and Delwiche (1991 b) were developed an aigorithm consisted of shading correction, segmentation of defect areas, edge extraction, and classification of the defect analysis and reported that the error rates for the N[R and colour system were 31% and 40% respectively. Using this system, they classified the defects or stern cavity into eight specific types (scar, stem cavity, Cut, bruise, scale, wormhole, brawn rot, and noise). In addditicrn, R.ehkugler and Throop (1989) employed line camera to capture NIR images and reported that digital imaging with statistical pattern recognition was found a useful technique for detecting bruises on apples, Furthermore, Throop and Aneshansley (1093) developed a method to detect both old and new bruises from NIR rcflectanee images of stored `Delicious' apples, Upchurch et al, (1990) investigated a spectrophotometer method for detecting bruises on apples and Yang (1994) introduced a flooding algorithm which is applicable to surface feature detection for apples or other types of fruit with relatively uniform skin color. Wen and Tao (1998) successfully developed a dual-camera based on NIRIMJIR imaging method for apple defect reeogniion and stemlcalyx identification and they reported that this rules-basil approach have more flexibility for changing or adding parameters, features and rules to meet various sorting requirements than the neural network method. Xing et al (2006) provided a method based on NLR imaging for bruise detection on Golden Delicious' apples and they studied the potential of NIR imaging in determining maturity of Along fora indkca L, In another research by Delwiche et al. (1983), it was found that the skin ground color was a better indicator of edible quality than flesh thmness, Jima et al. (2007) investigated bakery characteristics of wheat dough and of the final product that made by it using the N[R imaging techniques and reported that the potential. of NIR imaging technique can be used in the prediction of the protein content and zeleny sedimentation, also. The NIP. imaging; an indirect method, can also be applied in arable crops to detect and locate the insect, but, it cannot detect low level infestation in bulk samples between live and dead insects (Dowell et at 1999). For better results, Ridgway and Chambers (1998) captured NIR images of infested wheat samples by using a NIR vidicon camera and detected insects using comparison method and Davies et al. 39 (2003) checked the potential of NIR imaging for detecting and Iocating insects in cereal grain, and they reported that how line segment detection can be achieved with a minimum computation if two masks embodying a vectorial design strategy. Similarly, a near-infrared imaging system for determination of rice moisture was developed by Lin at al. (2006) and reported that the performance of NIR imaging system was almost the same as that of NIBS (near-infrared spectrometry). McClure (1987) determined moisture content in rice using NIR system. Therefore, NIR imaging system can oe used for prediction of various properties and also for detecting the defects, etc. But, the main drawback of NIR imaging system is it needs for frequent calibration as it is very sensitive to moisture content in samples (McClure, 1987). 2.3.1.4. Thermal Imaging Thermal imaging is a non-contact technique to convert the radiation pattern of an object into a visible image called a thermogram or thermal image (Agerskins, 1975). By this method, the surface temperature of any object can be mapped two dimensionally at a high resolution and this two-dimensional temperature mapping technique has Potential to characterize products during several operations in agricultural and food industries. Thermal imaging is a passive technique which does not require any external source of illumination. In thermal imaging, the invisible radiation pattern of an object is converted into a visible image and this makes easy to detect invisible defects on the object's surface. So far thermal imaging has been successfully adopted for studying plant physiology, irrigation scheduling, and yield forecasting in agricultural fields. Likewise, maturity evaluation, detection of bruises in fruits and vegetables, detection of spoilage in agricultural produces by microbial activities, and detection of foreign materials are the potential post-harvest operations to use thermal imaging (Manickavasagan et aI. 2005). Vanlinden et al (2003) used thermal imaging method to detect bruises in tomato, The thermal imaging is categorized as infrared and far- infrared imaging and the infrared thermal imaging system between these imaging is more suitable for making quality determination of agricultural produce on the basis of surface temperature than quantitative measurement (Davis and Lettington t988). There are several techniques such as x-ray tomography, infrared tIxermography; electrical impedance tomography, ultrasound imaging, microwave radiometry, and magnetic resonance imaging 40 (M I) are available to reap the temperatures of biological materials (Mott and Hall 1999, Sun at al.1993 and 1994. Kantt et al. 1997, 1998, Hulbert et al. 1995). However, the infrared thermal imaging has great potential for both pre-harvest and post-harvest operations in agriculture due to the portability of the equipment and simple operational procedure. The Legion in the infrared band with ravlepgt,hs frorn 3 to 14 im is called the thermal infrared region. Hellebrand et al. (2001) have reported that tEe thermal imaging can visualise and quantitatively determine the distributed surface temperature of horticultural prodace after a lot of testing and examinations of the quality changing pattern of horticultural produce by using the infrared vision. The thermal imaging is able to measure all those produce properties that connected with thermal processes (transpiration, respiration) and infrared imaging is now leading to the qualitative new insights and contributing towards quality rnainter ance. The unique application of thermal imaging is evaluation of maturity in both pre- and post-harvest stages of fruits and vegetables, which is not only a time consuming process but also affected by human fatigue {Darns et al. 198(I). Even though several automatic methods are available for this purpose, visual inspection followed by thermal imaging is more suoc ssful in many parts of the world. The maturity evaluation of the tomato, pear, and persimmon using infrared tlnermome r was done by Danno et al, (1980). The surface temperature of irnxnatnre fruits, stared at lower temperature was slightly higher than that of mature and over-ripe fruits. In contrast, the surface temperature of immature fruits when stored at higher temperature was found slightly lower than that of mature and over-ripe fruits. Conclusively, it can be stated that the surface temperature of fruits and vegetables were found different in all three grades of maturity (immature, mature and over-ripen) and this difference was reported by Davits et al. (1980) in the range of 0.5 to 1.O° . Similarly, Varith et al. (2003) also determined the capability of the thermal imaging to detect bruises in apple and tomatoes and reported that the temperature difference of tomatoes by heating artificially bruised tomatoes in microwave oven for 146 s and imaging thermally found between the intact and bruised parts is in the range of 0.5 to 1,0°C. Image capturing, therefore, is an application program that enables users to upload pictures from digital cameras or scanners which are either connected directly to the computer or to the network. A system consisting of a high-pixel resolution CCD (charge coupled 41 device) chip and associated hardware is the most common method for generating digital images. However, because digital images are inherently monochrome, or black and white, other hardware and software are needed to generate color images. An electronic system that can classify omits by mass, by colour, by diameter, by bruising, shape and size should be composed of a camera, a Sees, a light source, a filter, a PC and image processing software (Sarkar and Wolfe 1985; Hahn 2000). Lino et al (2008) used electronic systems composed of CCD camera and PC for image capturing in quality evaluation of tomatoes and lei ions. RecentLy Kanali et al. (1998) investigated the feasibility of using a charge simulation method (CSM) algorithm to process primary image features for three dimensional shapes recognition. However, only -dimensional (2D) data is needed for graditg, classification, and analysis of most agricultural images. 3- dimensional may be needed for information an structure or added details. Sonka et al. (1999) have developed 3-D vision technique from a series of 2-D images to derive a geometric description. The imaging unit consists of a high performance 1 -bit digital CCD camera and spectral range 290-1100 nm with the peak quantum efficiency of 60% at 500 nm were used for quality evaluation of pickling cucumber using hyperspectral reflectance and transmittance imaging by Ariana and Lu (2008). The process of converting pictorial images into numerical form is culled digitisation. In this process, an image is divided into a two dimensional grid of small regions containing picture elements defined as pixels by using a vision processor board called a digitiser or frame grabber. There are numerous types of analogue to digital converters (ADC) but for real time analyses a special type is required, this is known as a flash' ADC. Such `flash' devices require only nanoseconds to produce a result with 5(1-200 mega samples processed per second (Davies 1997), Selection of the frame grabber is based on the camera output, spatial and grey level resolutions required, and the processing capability of the processor board itself (Gunasekaran and Ding 1994). The detailed descriptions of other components of computer vision system have been given in next section 2.4. 42 .. Image processing tools and techniques used in computer vision system for food quality evaluation Various image processing techniques have been developed during the last decade for food quality evaluations and have found many applications in diverse fields of scientific, commercial, and technical endeavors, A considerable effort must therefore to be made to review and describe the tools and techi itlues that are used in computer vision to facilitate image processing. So, the aim of this paper is to review and investigate recent applications of image processing tools and techniques that are adoptable or have already been adopted by the food industry. The feasibility of various technitlues used in computer vision for food quality evahuation is also discussed. This review can serve as a foundation for applying the processing techniques available and also for the development of new processing techniques in computer vision systems. 2.4.1. Image processing In computer vision, grouping parts of a generalized image into units that are homogeneous with respect to one or more characteristics (or features) result in a segmented itnsge. A scented image extends the generalised image in a crucial respect; it contains the beginnings of domain- dependent interpretation. So. imaging is the representation or reproduction of an object's outward form espeiaIly a visual representation or formation of an image (Anon 2010). And an image. is a representation, likeness,-or imitation of an object or thing, a vivid or graphic description, something introduced to represent something else. 'Usually, it is a condensation or summary of the information of the object it represents. Ordinarily} an image contains less information than the original abject and is always incomplete, yet in some sense it is an adequate representation of the abject. Owing to imperfections of image acquisition systems, the images acquired are subject to various defects that will affect subsequent processing (heng and Sun 005b). The field of digital image processing for such corrections refers to image processing by means of a digital computer. A fundamental computer vision system generally includes the following functions: lighting (dedicated illumination), optics (to couple the image to a sensor), sensor (to convert optical image to an analog electronic signal), analog-to-digital (AID) converter (to sample and to quantize the analog signal), image processorlvision engine (which includes 43 software or hardware to reduce noise and to enhances process, and analyze an image), computer (decision-maker and controller), operator interface (terminal, light pea, touch panel display, and so on, used by operator to interface with the system), input-output (communication channels to the system and to process it), and display (television or computer monitor to make visual observations} (Zuech 2000). Among these functions, processing is the act of subjecting something to a process, a series of actions or operations leading to a desired result by altering its form in a desired manner that takes an image and makes it into an image, starting with one image and producing a modified version of that image. Digital image analysis is taken to mean a process that takes a digital image into something other than a digital image, such as a set of measurements of the object (contained by the image). Before processing, an image must be converted to its numerical form, which is called digitization. A digital image is composed of a finite number of elements, each of which having a particular location and value, referred to as picture elements, image elements, pels, and pixels. The term pixel is most widely used and denotes the elements of a digital image (Gonzalez and Woods 2002). The term digital image processing, however, is loosely used to cover both processing and analysis. Table 2.6 displays some important -commonly ud huge processing. terms. 2.4.2. Image processing techrd que5 Image processing techniques generally consists of the following 5 steps, also shown in Fig, 2..óa, which are (1) image acquisition operations to convert images into digital form; (2) preprocessing operations to obtain and improve an image with the same dimensions as the original image; (3) image segmentation operations to partition a digital image into disjoint and non-overlapping regions; (4) abject measurement operations to measure the chraeteristies of objects, such as size, shape, color, and texture; and tS) classification operations to identify objects by classifying them into different groups (Du and Sun 2004a). Fig. 2.6b, however, shows different steps of image processing techniques in terms of lbw-, mid- and high-level. Image Pre- Image Object acquisition Processing Segmentation Nleasurcraent Iassification Fig. 2.6a. Configuration of common image processing system u and Sun 2004a) 44 Table 2.6. Commonly used image processing terms Terms Ileflnitlun Algorithm An algorithm is any program the user uploads that can be- used to process the images, extract features, or classify an image. Binary image Image where pixels have only two values, generally 0 and 1. Brightness The gray Ievel value of a pixel within an image that corresponds to energy intensity. The larger the gray level value, the greater the brightness. Closing A dilation followed by an erosion. A closing fills small holes in objects and smoothes the boundaries of objects. Contrast The amount of gray level variation within an image. Dilation A morphological operation that enlarges the geometrical size of objects within an image. Erosion A morphological operation that reduces the geometrical size of objects within an image. Gams correction Gamma correction allows users to better match the intensity of their prints to what they see on their computer screen (CRT). Gray scale Range of gray shades, or gray levels corresponding to pixel values that a rnanochrome image incorporates. Histogram A graph showing the number of pixels in an image at each different intensity value found in that image. Image A two-dimensional function, f(x, y), where x and y are spatial (plane) coordinates, and the amplitude of f at any pair of coordinates (x, y) is called the intensity or gray level of the image at that point. Image Analysis Image analysis is the extraction of meaningful information from images; mainly from digital images by means of digital image processing techniques. Image processing Encompasses various processes and analysis functions that we can apply to an image. Image preprocessing The pre-processing task involves some procedures to prepare the "images to be ready for image processing. Imaging Any process of acquiring and displaying images and analyzing image data. Kernel A kernel is a (usually) smallish matrix of numbers that is used in image convolutions. mask Refcc to a small image used to specify the area of operation to take place an a larger image in an algorithm. Matrix Image, representation using MxN matrix is a 90° clockwise rotation of the conventional two-distrensional Cartsiatt coordinate representation. Morphology Originally comes from the study of forms, of plants or animals. Image morphology represents study of topology or structure of objects from their images. Morphological processing refers to certain operations where an object is "hit" with a structuring element and thereby reduced to a more revealing shape. Noise Degradation of image due to equipment (such as sensor, camera misfocus), type of modality, motion, turbulence, and sa on. Opening An erosion followed by a dilation. An opening removes small objects and smoetlies boundaries of objects in the image. Pixel Slang for picture element, the smallest element if an image. Preprocessing Noise reduction, image enhancement, and so on. Semeatatioa Partitioning an image into objects of interest Thresh lding A value used to segment the gray level values of an 'map into two different regions. Also called tho bin arization of an image. 46 2.4.2.1. Law level image techniques Low-level processing includes image acquisition and preprocessing. Image acquisition Image acquisition, the first step in any image processing system, is the capture of an image in digital form. It is the transfer of the electronic signal from the sensing device to numeric form. Its precision generally depends upon factors such as specification of illumination system, quantum efficiency and spectral range of eamera'sensor, illumination techniques, and field of application (Martin 200 ). Among these, illumination is an important prorequisile in image acquisition for food quality evaluation, since the -performance of the illumination system can greatly influence the image quality and play an important role in the overall efficiency and accuracy of the system (Novini 1995). A wide variety of light sources and lighting arraagernents are, however, available (Tao et at 1995), a well-designed illumination system can help to improve the success of the image processing and analysis by enhancing image contrast (Novini 1990; Sun 2000). During the last decades, cornsioer ble research effort has been directed at developing techniques for image acquisition, Avery intensive field of research in image acquisition is the development of sensors. Widely various configurations of sensors have been used to convert images into digital form, The sensors such as charge-coupled device (COD) and CMOS (complementary metal oxide silicon) camera are widely used to obtain images of food products. The CCD camera is frequently employed by image processing systems for food quality evaluations. It has been viideiy used for quality classification, physical characteristic detection, and property estimation of food products. On the other hand, CMOS image sensors have intrinsic advantages (low power can.imptiou1 low cost, high speed imaging, integration capability, radiation hardness, etc,) that make them well suited not only for low-cost imaging markets but also fo.r high performances applications such as, high end Digital Still Photography, high-Definition Television and several space applications. These sensors are two main kinds of solid state image sensors which are widely applied at present. CCD and CMOS image sensors used for photons detection are organized as arrays of photo detectors that deliver an electrical signal reIated to the amount of photons that fall on the pixel surface during the integration time. 47 Lighti 0b+ et Arran eme Camera setup linage acquisition Pre- processing Low level Image processing Contrast Noise enhancement . removal I Scgmeniation Threshold I I Region based Gradient Classification Hybrid based based segmentation based based segrncatatiou segmentation segmentation J segmentation Interrnedia to level Image I Representation processing Description Recognition high Ievel Image i1ei'tetdttori processing Results Fig.2.6b: aciaus steps of low-, mid- and high-level image processing techniques 48 They. both use the photoelectric effect in silicon, in either I photo gate or a photadiode detector (Zhang et al. 2008). Since, it is difficult to evaluate food quality in the ordinary spectral region with a single camem, the CCD and CMOS image sensors will remain complete&ntary and Iong term competition, and flourish the image sensor market together (bang et al. 2008) in predictable future. For example, Park et aL (1998) implcmcnted a multi-spectral' imaging technique with 4CCD cameras in an on-line inspection system to separate wholesome and unwholesome chicken carcasses. Li et al. (2002) developed a novel automated experimental system for sorting apple surfaeo defects based on computer image technology ht their study they mounted above and below the cwiveyo' 2 C I) cameras, with interference Land-pas-s optical filter (840, urn). Omid vt al. (2010) developed a machine vision system consisted of two CC]) cameras for computing the volume and mass of citrus fruits. So, to achieve a basically complete inspection of food products, it is necessary to use, more than one eawer4 to obtain food images free of geometric distortion from different directions. In spite of this, images are often degraded because of distortion and noise in the camera and the optical system. So, to preserve image quality reasonably enough for valuable information retrieval before the beginning of analysis and processing, digital images must be preprocessed. By Preprocessing the image is enhanced. Image preprocessing refers to The initial processing of the raw image data using oprations such as noise reduction, contrast enha.ncemmt, smoothing, image sharpening, transformation (converting the original 3D image to 21) pixel values), correction for blurring and fioeusing, correction of geometric distortions, and gray level correction, etc. (Shirai 1987; Sonka et al, 1999; Sun 2000; Panigrabi et aL 2001;Gonzales et al, 004). Faucitano et al. (2005) used the free hand mask cutting tool of Corel Photo-Paint 8.2 (Coral Cooperation, nada), an image editing pro -am to cut out the muscle image with no background in their study of pork matbling characteristics. Noise removal An image can have different types of noise due to the various means of image capturing ystenis, such as read-out noise, wiring noise, electronic noise, and extraneous 49 noise like as periodic stripes introduced during the digitization process. All these noises must be removed for good image quality; and by suppressing undesired distortions, or by enhantiz~g irmpoitant features of interest before digitizing and storing the images in computers that can be possible. Averaging and Gaussian filters are often used for noise reduction with their operation causing a smoothing in the image but having the effect of blurring the edges. In spite of this, the most efficient and feasible approach for image noise removal is averaging the iinae by itself (Zheng and Sun 2008a) Yarn and Papadakis (2004) averaged values for multiple pixels to reduce the noise in the, plats d ring the measuring and analyzing the color of food surfaces. The simplest method of averaging an image by itself is the Iinear filter. By linear filter the intensity values of pixels, within the small region of an image, are averaged using the intensity values of their neighboring pixels. The weighting and size of the filter can be adjusted to remove different types of noise (Zheng and Sun 2008a). Generally, linear filters include a high-pass filter (Gradient and Laplacian) and a low pass filter (smoothing and Gsion. Median filter, a nonlinear filtering technique, is another popular filter. Like the low- pass filtering. median filtering smoothes the image and is thus useful in reducing noise. Un ike low-pass filtering, median filtering can preserve discontinuities in a step function and can smooth a few pixels whose values differ significantly from their surroundings without affecting the other pixels. Median filter is often applied to gray value images due to its property of edge preserving smoothing (Kosechan and Abidi, 2001). It is also known as bilateral filter, originally proposed by Tomasi and Manduchi (199), was employed by Du et al. (2008) as a heuristic tool for noise removal in beef images. Du and Sun (2004) used a median filtering method to remove possible noises within a pizza base image and reported that the median filter replaced the output pixel with the median of its neighboring pixel values instead of a weighted sum of those values that helped in the thresholding-based segmentation of pizza images. Labbafi et al. (2007) applied a median filter, to remove impulse noise and to close the interfaces thereby preserving the edges and contours, and developed an on-line optical method for the assessment of the, bubble size and the morphology of aerated food products. so There are 2 primary advantages of great use of the median filter in the food industry (Du and Sun 2004a, 2006a; Faucitano et aL 2005): it does not shift the edgei of an image, as may occur with a Iinear filter (Russ 1999), and noise removal does not reduce the difference in brightness of images. In addition, this technique can b-a considered as a special case for filters called sank statistic filters, allowing the edges to be preserved while f ltering out the peak noise. For that reason, the median filter is often used before applying an edge detection technique. Leernans et al. (1998) used 2 types of filters: a 3x3 median filter' and a `3x3 box filter' for segmenting defects on 'Golden Delicious' apples to preserve the main apple defect as much as possible. Goodrum and Elster (1992) applied the °filter facter', namely, the modified unsharp filter transform, which is a Laplace transform of an image added to the same image, to enhance cracks in an egg image without overly enhancing other surface features and noise. Bennedsen em al. (2005) identified defects in images of rotating apples using images taken with 740- and 950-nm filters to eliminate false positives reportedly caused by shadows; in earlier investigations seis of interferometric filters had been applied to analyze variations in light intensity, as recorded by a black-and-white video camera. Thus, a filter can be used to discard those images which add little or no infxrmation to the system icNitt-Gray et al. 1995) and to focus the attention of the machine vision system on the same phenomenon being visually observed. Contrast enhancing Contrast generally refers to the difference in luminance, or grey level values, in an image and is a most important characteristic of an image. Sometimes images captured are of low contrast: the intensity values of the images are within a small range of intensity levels, and thus pixels with different intensity values are not well distinguished from each other. The process to irmerease the difference in intensity values among pixels, so that they can be effortlessly distinguished'by human or computer vision is known as contrast enhancing. Two definitions have been commonly used for measuring the contrast of the test targets. One is Michelson formula [Eel. (1)1, used to measure contrast of a periodic pattern such as a sinusoidal grating and other is Weber fraction [Eq. (2)1, used to measure the local contrast of a single target of uniform luminance seen against a uniform background. 51 L --L max min C = /L (2) where L and L. are the maximum and minimumi 1iin inance values, respectively, in the gratings and AL is the increment or decrement in the target luminance from the uniform background luminance L (Deli 1990). The contrast ratio has a strong bearing on the resolving power. The larger this ratio, the more easy it is to interpret the image (Bagade and Shandilya 2011). Histogram scaling and -equal.ization are 2 popular contrast enhancement techniques. Histogram equalization is the most utilizedd technique in the food industry for contrast enhancing of images (Jain t99). In this technique, a histogram of the original image is redistrihrtted to produce a uniform population density, which is obtained by grouping certain adjacent grey values and by spreading out the number of pixels at the histogram peaks, thereby selectively compressing those at the at the histogram valleys (Gauch 1992; Bagade and Shandilya 2011): In contrast, in histogram scaling the original histogram is transferred from one scale to another (mostly from smaller to larger). The histogram scaling function can be linear or nonlinear and also can be one-to-one or multiple-to-one, Most of the transform functions for histogram scaling are limited to proposed cases. Histogram equalization is operated on an image in 3 steps; histogram formation, new intensity value calculations for each intensity Level, and replacement of the previous intensity values with the new intensity values (Bagade and Shandilya 201 l). A contrast limited adaptive histogram equalization method was developed by Du and Sun (2006a) and applied to adjust pork images by facilitating the segmentation of pores. In this method, they enhanced the contrast of the images by dividing each image into nonoverlapping small regions and rhea enhancing the contrast in each small region. So, by enhancing image corltrast, the accumcy of the image processing system could be improved and lead to success of image analysis (Ciunas ekaran 1996). 2.4..2. Mid-level processing- segmentation techniques Mid-level processing involves image segmentation and image representation and description. Segmentation of Food images, which refers to automatic. recognition of food 52 products in images, is of course required after image acquisition, because food quality eva1uaticn is completely and automatically conducted by computer programs, without any human participation in computer vision techniques (heng and Stan 2(JO$a). The segmentation techniques allow partitioning of images into regions that have a strong correlation with objets or areas of interest, as shown in Figure $ (Sun 2G00). As the subsequent extracted data are highly óependent on the accuracy of image segmentation, this operation is one of the most important steps in the entire image processing technique (onka et al. 1999; Sun 2000) and also most difficult tasks in image processing (Harrabi and Ban Braick, 2011). A large number of segmentation techniques have been developed. Of these, thresholding-based, region—based, grrdient-based, and c1assilcation-based segmentations are the 4 most popular techniques in the food industry (Zheng and Sun 2008a). However, until now there is no general technique that can solve all the different image segmentation type (Harrabi and Ben Braiek, 201 I). ThreshhoIdhs-based segmentation Thresholding is a technique frequently applied to image segmentation and widely used in many image processing applications such as medical image processing tYan et a. 2005), detevtic11 of video Change (Sing et al. 2-005), optical color recognition (Shaba and Udupa, 2001), food processing, etc. It chooses proper threshold u T to divide the image pixels into several classes and separate the objects from background (Kaur and Kaur, 2011). However, the fundamental principle of threghvlding technique is based on the characteristics of image. This technique -art be divided into bi-level and multilevel category. In bi-level thresholding, a threshold is determined to segment the image into two brightness regions which correspond to background and objects while in multilevel thresholding, more than on threshold will be determined to segment the image into certain brightness regions which correspond to one background and several objects (Huang et al 2005). Several methods have beets proposed to automatically select the threshold such as two dimensional entropy. (Abutnleb 1989), through moment preservation technique (Tsai, 1985) maximizing of nonfuzziness of the 2D grayscale histogram (Wang et al., 2002), discriminant analysis or variance based (Otsu 1979), etc. used in bi-level thresholding. The methods of threshnldiog based on distribution function (Buukhrouba et al, 1985), quad-tree approach which combines 53 THESIS both statistical and spatial information (Spann and Wilson, 1985), hill-clustering (Paparnarkose and Gato e, 1994), connectivity preservation criteria (O'Gorman, 1994), verification based multi threshold probing scleme'(JJang and Mojon, 2003), etc. are used in multilevel thresholding. Further, thresholding can be divided into global, local and dynamic thresholding tech ques (Kaur ad )faux, 2011). However, only global threshold (fixed value) is used in practice, known as bi-Ievel and tri-level thresholding. The 4 soma methods for the selection of the global thxesholding are manual, isodata (Rider and Calvard 1978), objective function, and histogram clustering. Among these, manual selection is not ideal for online automatic food quality evaluations using computer vision: Ridier and Calvard (1978) developed the first automatic threshold selecting method, namely the isodata algorithm. The objective function method can be used alternatively. Variance-based (Otsu 1979) and entropy-based (Pan 1980; Kapur et al. 1985; Sahoo et al. 1997) are the 2 kinds of objective functions that are used mostly. Among these, Otsu's method is referred to as one of the most powerful methods for Si-level thresholding application (Sahoo et al., 1988). Otsu method can provide satisfactory results even in the case of unirnodel histogram images which do not have two obvious peaks and can be used as a classical method in real thresholding application (Can et aL, 2002). For gray level &-ay level image (x, y), Otsu (or bi-level) thresholding is used to transform (x, y) to binary image g(x, y) by a threshold T which can be expressed as: 0, iff x,y) 7T ~1, otherwise However, using the Otsu's technique alone uneven lighting gray level images cannot be thresholded accurately and effectively. Huang et al. (2005) have a method for thresholding in partitioned windows based on pyramid data structure manipulation with window size adoptively selected according to Lorentz information measure to solve the probl s'n of uneven lighting and reported that to emphasize the partitioned windows technique, only Otsu's thresholding method was considered. Since. Otsu's technique can be easily extended to other bi-level and multilevel thresholding. So, multilevel thresholding can be considered an extension of hi-level threshoIding in which a gray level image /(x, y) is transformed to a muftiIevel image g(x,y), by several thresholds T, , T ...... , 7',~ , as: 54 C. tNlo.....u..,....-w+ 0 if .f (x, v) T, 1 if 7, f(x, y) : T2 &'Y)= m if f(x,y)' T'R" Clustering is an example of multiclass thresholding and the methods entropy, metric, moments, and interclass variances are reserved for strictly binary thresholding techniques. Based on bi-level thresholding method, Teoh and Syaifudin (2007) discriminated mango object from the background offltered mango image during processing and analysis of image for estimating weight of Chokanman mango. Bulanon et al (2001) used multivariable thresholding to develop a machine vision system to guide the developed robotic hand and found that 80% of the apples segmented using two color models (LCD and chromaticity). A new method 'reduced—dimension clustering (RD C)! was developed by Steward et al. (2004) for segmentation of vegetation and reported that the segriieritation pertbrmance was consistently high, with average segmentation success rates of 896% and 91,9°% across both cloudy and sunny lighting conditions, respectively. h LJ4i: Fig.2..7. Typical segmentation techniques: (a) thresholding based segmentation (b) edge based segmentation (c) region-based segmentation (Sun 2000) In addition, Panigrahi et al, (1995) have developed an automatic thresholding technique to segment the background from the images of corn germplasm and found that the modified Otsu algorithm performed better than the no modified Otsu algorithm. The variance-based objective function generally performs better than the entropy-based one, except for images in which the population of one class is relatively larger than that of the other (Read 1982). But when the population of one class over the other is lower than O.01, 55 the variance-based objective function produces erroneous results (Kittim and Yilingworth 19S6). The histogram clustering (k means clustering) method is mainly used in threshold selection. In this method, an intensity value from Ito Lis picked as the threshold to segment the histogram into 2 classes, object and background, with mean intensity values (beng and Sun 2008a). In spite of the techniques described above, other thresholding-based techniques used are the window extension method (Hwang et aL 1997), and fuzzy thresholdirig techn?que (Tobias and Seam 2002). Sometimes, only the thresholding technique is not enough to segment an image becatise t e contrast of objects varies within the image. So. other techniques can be used to convert the image first and then segment the processed image with thresholding. Based on the reference image of an apple, Li and Wang (1999) developed a method to accomplish defect segmentation for a curved fruit image. In their work, they generated a reference fruit image and normalized it to achieve the original fruit image for inspection. Subtracting the normalized original fruit image from the normalized reference fruit image an image was obtained, and then they extracted defects by applying a simple thresholditg method. However, it was difficult to detect defects in apple surfaces using any simple global threshold segmentation algorithm, except for adaptive thresholding methods (Ti et al. 2002). Region-based segmentadon Region-based segmentation methods are usually proposed for segmenting complex images in which the number of classes is large and unknown. However, region-based methods are less popular in the applications of computer vision in the food industry. Generally, there are 2 region based-segmentation techniques: growing-and-merging (GM), and splitting-anthmerging (SM) (Navon et at. 2005), In GIB, neig iboring pixels iteratively are merged into the pixel, selected as a growing region, until no more pixels can be merged, Afterwards, the growing procedure is repeated with an-other pixel that has not been merged into any regions, ur it all the pixels in the image have been merged into various regions to find out regions that are too small to remain as independent, mostly due to the presence of noise. In other words G. !! is a bottom-up method that groups pixels or sub-regiolls into larger regions according to a set of homogeneity criteria (Du and Sun 2004a). Sun and Du (2Q04' proposed a GM algorithm for the se mentatton of pizza toppings which is impossible by thresholding-based methods. They reported that the region-based segmentation methods are less popular in the application of computer vision in the food industry and have limited instances ofuse, except GM. In the SM method, the whole image is initially regarded as a big region, and is split iteratively into smaller regions with uniform image characteristics (such as color, gradient, and texture). In other words, SM is a top-down method that successively divides an image into smaller and smaller regions until certain criteria are satisfied (Du and Sun 2004a) . Yang (1994) developed a flooding algorithm, a region-based segmentation method, to detect apple surface features by introducing the concept of topographic representation. In their sludy they treated detection of the patch-like features as one of catchment-basin detection in apple grey-level landscapes and reported that, after the flooding process, the catchment basins became lakes for which geometric parameters such as area and perimeter can be easily extracted. Sun (2O0O) developed a new region-based segmentation algorithm for processing pizza images. It employs the traditional region-based segmentation as a dominant method and combines the strengths of both thresholding and edge-based segmentation techniques. This new algorithm adopted a scan-line-based growing mode instead of the radial growing mode employed in a traditional region growing algorithms. First, it partitioned a pizza image into horizontal or vertical lines after edge detection and then merged the lines into small homogeneous regions. Finally, the small regions were merged into larger regions that represent topping objects. Fig, 2.8 shows an examp]c of pizza images and its segmentation result using the new region-based segmentation technique developed. ice L'• Fig. 2.9. Pizza images: (a) original image; (b) segmerded image (Sun, 2000) 57 Region-based algorithms are computationally more expensive than the simpler techniques, such as thresholding-based segmentation, but region-lased segmentation is able to utilize several image properties directly and simultaneously in determining the final boundary location. It shows the greatest promise in the segmentation of food products because strong prior knowledge is not available. Gradk,i- basedsegrnentation The thresholding approach accomplishes segmentation by partitioning the image into sets of interior and exterior points. By contrast, gradient-based approaches attempt to find the edges directly by their high gradient magnitudes, Gradient-based segmentation method is similar to edge detection based on the gradient of an image (Du and Sun 2004b). Computing the image gradient is favored simply because boundaries of local contrast can be effortlessly observed in the gradient images, thus edges of the objects can also be easily detected, Image segmentation is meanwhile accomplished, since the edges of objects in images are located. Therefore, gradient-based segmentation is also called "edge detection_" Edge-based segmentation relies on edge detection by convolute edge operators (Jain 1989). Considering the image as a function f of the intensity value of pixels (x, v), the gradient g can be computed by: (2) = KJ1 c;j2 In a digital image, a gradient operator is similar to an averaging operator (for noise removal). Some of the well-known gradient operators that have been widely used are Sabel, Prewitt, Roberts, and Kirsch operators (Russ 1999). Although most of these operators are competent when the intensity transition in images is very abrupt; and as the intensity transition range gradually gets wider and wider, the gradient operators might not be as effective as they are supposed to be. So, the second-order derivative operators are depicted as alternative approaches by researchers far gradient operators as 72 (3) The Laplace operator is one of the more widely used derivative operators in which the second- order derivative is determined by subtracting intensity values of the Aeighboring 58 pixels from the value of the central pixel (Marr and Hildreth 1980). However, these operators have not yet seen employed in the food industry. The Canny edge detector, widely used in the food industry for boundary extraction of food products, performs maximum suppression of false responses including the assessment of the optimal signa1.to-signal and optimal locality (Canny 1986). The active coatour model (ACM) is another popular gradient based technique and is known as "Snakes," which trap forms the problem of edge detection into an energy optimization (Zheng and Sun 2008a). The ACM method can be used for the segmentation of touching adjacent rice kernels (Wang and Chou 1996). Jia et al. (1996) used image segmentation algorithms involving edge detection and boundary labeling and tracking to locate the position of whale fish. A Canny edge detector with Gaussian smoothing parameter 1,0 was selected to obtain the fish edge image and a labeling and tracking algorithm based on a recursive procedure was developed for locating, tracking, and thinning the fish boundary. However, the application of the gradient based segmentation is l n itetid because completed boundaries are di ic.ttlt and sometimes impossible to trace in most food images. Edge operators detect discontinuities in grey level, coIor, texture, and so on. Region segmentation involves the grouping together of similar pixels to farm regions representing single objects within the image. ttssIf ic&&ion-based segrrreilarion Classification-based segmentation is the second most popular method, after thresholding-based segmentation, used in the food industry. It is a pixel-oriented and each pixel is regarded as being an independent observer whose variables are generally obtained by image features (such as color, shape, and texture). Afterwards, a matrix that contains every pixel as an observer is obtained as the input forclassification. Each observer is then classified (object and background, or defect and nondefect, and so on) according to its variables, using a learning model (J u and Sun 2006b). Furthermore, features like extraction technique have drawn strong • interest from researchers carrying out work on apple quality evaluations wing computer vision technology ( ieynen et at. 2005). Sometimes, to acquire more information about a pixel, its feature can be extracted from a small region that is centered on the pixel. Therefore, besides the intensity value, the image texture, which is an important factor of the product surface for pattern 59 recognition due to its powerful discrimination ability (Arnadasun and King 1989), can also be extracted as a classification feature of pixels. For further technical information on the extraction of image textr a features, refer to the review by Zheng et al. (2006a). Since a large amount of data is present in the input matrix for classification, it is generally preferred that the dimension of the originalmatrix is reduced before classification. Accordingly, a self- organizing map (SOM) has been developed and generalized by extracting the intrinsic topological strczeture of the input matte from the regu rizations and correlations among observers♦ Afterwards, the SOM can be used for classification rather than the original observers, and the observers are assigned to the class of the neuron that the observers belong to (Cbtioui et aI. 2003; Marique et al, 2005). Classification-based methods attempt to assign each pixel to different objects based on classification techniques like statistical, fuzzy logic, and neural network methods (Du and Sun 2004b). A Bayesian classification process was used successfully to segment apple defects Leernans et al. 1999). In this process, the color frequency distributions of the healthy tissue and of the defects were used to estimate the probability distribution of each class. The results showed that most defects could be segmented by this rmethod, airhotigh russet was sometimes confused with the transition area between ground and blush color. In addition to this, fuzzy clustering can be used to simulate the experience of complex human decisions and uncertain information (Chtiot i et al. 2003; Du and Sun 2006b). In spite of the above discussion, an important caveat is that classification -based segmentation methods lack a matured way for coping with variations in rotation and scale, which limits their applicability(Du and Sun 2004b). Other Seg)StcfltUiiDlz technique. Although a large number of segmentation techniques have been developed to date, no universal method can perform with ideal efficiency and accuracy across the infinite diversity of imagery (Bhanu el al. 1995). So, we must develop and -combine several techniques in order to improve the segmentation results and increase adoptability of the methods. Watershed and hybrid-based segmentation techniques are more popular. For instance, Jiatem et al. (2003) developed an algorithm with an accuracy of 83% for the segmentation of cartilage and bone in images of vertebrate animals by using a thresholding-based method 6D twice, The concept of watersheds, which are introduced into digital images for rnorphological processing, originally comes from topography. The watersheds can be constructed from different scales of images grayscale (Viacerjt and Snipe 1991), bury ( anent et aL 2001), and gradient (Du and Sun 2006a). Owing to the presence of noise and local irregularities, there are far more minima from which far more catchment basins are formed, causing an over-segmentation of images. To overcame this problem, algorithms are assigned. One method for preventing over-segmentation of images is to eliminate the undesired minima, using morphological operators such as opening and closing. One such method was proposed by Du and Sun (2006a) to segment pores in pork ham images. In other methods, post-processing is conducted to merge over-segmented regions with similar image characteristics. Such a method with a graphic algorithm to determine the similarity of merging neighboring regions was developed by Navon at al. (2005). Despite all this, because image segmentation is by nature still an ill-defined problem, none of the methods desribcd can perform ideally across diverse irriages. it has been suggested recettly that several techniques rmight be combined together for the sake of improving the segmentation result and simultaneously increasing segmentation speed (Zion and Sun 200a). 2.4.2.3. High-level processing: image analysis After image s rnentation, where objects axe discriminated from the background, the characteristics of the objects, known as object maasmements, are calGuinted. These measurements are the core elements in a computer vision system, because they contain useful information for image understanding and interpretation and for object classification (Ballard and Brown 1982). Interaction with a knowledge database at all stages of the entire praeess is essential for more precise decision-making and is seen as an integral part of the image processing process. The operation and effectiveness of intelligent decision making is based on the provision of a complete knowledge base rliich in maces vision is incorporated into the computer. Algorithms such as new-al networks, fuzzy logic, and genetic algorithms are some of the techniques of building knowledge bases into computer structures. Such algorithms involve irnage understanding and decision-making capacities, thus providing system control capabilities. Neural network and fuzzy logic operations have been 61 implemented successfully with computer vision in the food industry (Ying et al. 2003). High-level processing involves recognition and interpretation, typicaLly using statistical classifiers or multilayer neural networks of the region of interest. These steps provide the information necessary for the process/machine control for quality sorting and grading (Brosnan and Sun 2004). For instance, measurements such as si , shape, color, and texture are carrying direct information and can be used in quality evaluations and inspection tasks. These measurements are rated as the primary object measurements, can be acquired from images {Du and Sun 2004b). Although a number of methods have been developed over the past few decades, there is not yet a perfect method for each type of measurement, and especially for texture measurements because of the lack of a formal and scientific definition of image texture while facing infinite diversity of texture patterns (heng et al. 2OO6a), 2.41.4. Other techniques Edge-demo on techniq ws Edge detection is an essential tool for machine vision and image processing. In image processing, an edge is the boundary between an object and its background. They represent the frontier for single objects. Therefore, if the edges of object's image can be identified with precision, all the objects can be lacnted and their properties such as area, perimeter, shape, etc, can be calculated (Bolhouse 1997). It is also the process of locating edge pixels and increasing the contrast between the edges and the background i.e. edge enhancement) in such a way that edges become more visible. In addition, edge, tracing is another terminology used by researcher. includes the process of following the edges, usually collecting the edge pixels into a list (Parker 1997). Some of the well-known edge detectors that have been widely used are the Sobel, Prewitt, Roberts, and Kirsch detectors (Russ 1999). The first quantitative measurements of the performance of edge detectors, including the assessment of the optimal signal-to-noise ratio and the optimal locality, the maximum suppression of false response, were performed by Canny (1986), who also proposed an edge detector taking into account all three of these measurements. The Canny edge detector was used in the food industry for boundary extraction of food products (Du and Sun 2004b, 2O06b; Jia et al. 1996). Digital nwrplwlogy 6Z Digital morphology is a group of mathematical operations that can be applied to the see of pixels to enhance or highlight specific aspects of the shape so that they can be counted or recognized (Parker 1997). In morphological processing, images are represented as topographical surfaces on which the elevation of each point is assigned as the intensity value of the corresponding pixels (Vincent and Soule 1991). One such method was proposed by Du and Sun (20(6a) to segment pores in pork ham images. In other methods, post processing is condicted to merge the over segmented regions with similar image characteristics together again. Such a method with a graphic algorithm to deterraine the simitai ity of merging neighbouring regions was developed by Navon et al. (2005). Texture effectively describes the properties of elements constituting the object surface, thus the texture mCasurements are believed to contain substantial information for the pattern recognition of objects (Arnad an and Ring 1989). The repetition of a pattern or patterns over a region is called texturo, This pattern may be repeated exactly, or as set or small variations, Texture has a conflictive random aspect: the size, shape, color, and orientation of the elements of the pattern (rarta i) (Parker 1994). Although texture can be roughly defined as the combination of some innate image properties, including fineness, coarseness, smoothness, granulation, randomness, lineation, hummock, etc., a strictly scientific definition has stiii not been determined (Hrslick 1979). .Accordingly, there is no ideal method for measuring textures, Nevert Bless, a great number of methods have been developed, and these are ategorized into statistical, siructtiral, transform-based, and model- based methods (Zheiig et al 2006a). These methods capture texture measurements in two different ways - by the variation of across pixels and their neighbouring pixels (Bliarti et at 2004). Thinning and skelthmizalion algorithms An image skeleton is a powerful analog concept that may be employed for the analysis and description of shapes in binary images. It plays a ceitral role in the pre- processing of image data (Gunasekaren 2408). A comprehensive review on thinning methodologies has been presented by .La n et al. (1992), In general, a skeleton may be defined as a connected set of medial lines along the limbs of a figure and skeletonization is a B3 process to describe the gIobal properties of an object and to reduce the original image into n more compact?representation. A basic method for skeletonization is thinning (Parker 1994). Ni and vnasekaran (1998) have applied a sequential thinning algorithm for evaluating cheese shred morphology when they are touching and overlapping. Naccache and Shinglial (1984) compared the results of fourteen skeletonization algorithm and Davies (1997) deduced the ideal shape after analysed the skeleton shape of object. Furthermore, image processing can be done independently by image acquisition software (Licbtenthaler et aL 1996). Mehl et al. (2004) have reported about, numerous, general and more specific image processing softwares/tools in his literature, Bailey et a1. (2004) demonstrated an image processing approach which estimated the mass of agricultural .products. rapidly and acutely. Koc (2007) determined the volume of watermelons and tashidi et al. (2007) estimated volume of kiwifruits using image processing. The surface area and volume of asyrn.metric agricultural products were measured by Sabliov et al (2002) using an image processing algorithm. Image processing algorithms are the basis for mmaebirie vision (Martin and Tosunoglu) and the image algebra (Ritter and Wilson 1996) forms a solid theareti a1 foundation to implement computer vision and image processing algorithms, A quadratic diserirniriant model based on an aIgorithm was developed by Harrell (1991) for consultation duri g on-line sortiig. Iipage analysis uses digital data to display images and to mathematically manipulate images to highlight various characteristics. Areas of homogeacous color or edges between difl<'erunt color areas are examples of useful image features, Often, filtering of the original data can help to eliminate image noise, such as shading caused by uneven fighting. Then, statistical and math niatical techniques are applied to the data to dis nguish elements of the image (Ballerd and Brown 1982 .Various operations and techniques are applied to the processed image to extract the desired information, Among these operations and techniques are, Object recognition-, `eature extraction; Analysis of position, size, orientation, etc- these terms being self explanatory in nature. A first order statistics which are based on statistical properties of the gray level histogram of an image region has been given in Table 2.7. 3-D Teelrnigrre in general, only 2-dimensional 2D) data are needed for grading, classification, and 64 analysis of most agricultural images. However, in many applications 3-dimensional image analysis maybe nceded as information on structure or added detail is mired. A 3-fl vision technique has been developed to derive, a geometric description from a series of 2-D images (Sonka et al. 1999). lu practice this technique unight be useful for food inspection. For example, when studying the shape features of a piece of bakery, it is necessary to take 2-D images vertically and horizontally to obtain its roux Miss and thickness, respectively. Table 1.7: First order statistics which are based on statistical properties of the gray level histogram of an image region Texure Features Expression Measure of Texture Average Gray ft ` H) A measure of average intensity Level 9 Standard Deviation A measure of average contrast Smoothness Measures the relative smoothness of the intensity in a region f3 is 0 for a region of constant intensity and approaches 1 for regions with large excursions in the values of its intensity levels. Skewness Measures the skewness of an histogram. f4 = —f1 ) 9 This measure is 0 for symmetric histograms, positive by -histograms skewed to the right and negative for histograms skewed to the left Uni faruAty r tt I (s} Measures uniformity. This measure is • 9 maximum when all gray levels are equal (maximal uniform) and decreases. from then. Entropy ff = — F1(g)kg. , (g) A measure of randomness Recently Kanali et al. (1998) investigated the feasibility of using a charge simulation method (CSM) agorithrn to process primary image features for three-dimensional shape recognition. The required features were transferred to a retina model identical to the prototype artificial retina and were compressed using the CSM by computing output signals at work cells located in the retina. An overall classification rate of 94% was obtained when V the prototype artificial retina discriminated between distinct shapes of oranges for the 100 data sets tested. uiwkaanend Di (19 9) obtained 3-D images of fat globules in cheddar cheese from 2-11 images. This ermbled the in situ 3-1) evaluation of fat glohule characteristics so as the process parameters and fat levels may be changed to achieve the Tequired textural qualities. 2.4.3. Applications of image processing tools and techniques used is computer vision system for quality evaluation of food and agricultural products 1.4.3-1. Assessment of fruits C. mputer vision has been widely used for the inspection and grading of fruits. It offers the potential to automate manual grading practices and thus to standardize teehniques and eliminate tedious inspection tasks. KanJi et a), (1998) reported that the automated inspection of produce using machine vision not only results in labour savings, but can also. improve inspection objectivity. Computer vision leas been used for such tasks as shape classification, defeats detection, quality grading and variety classification of the apple, Paulus and Solireveris (1999) developed an image processing algorithm based on Fourier expansion to characterize objectively the apple shape so as to identify different phenotypes. E,cperimeritatian by Paulus et al. (1997) also used Fourier analysis of apple peripheries as a quality insp tion/classif cation technique. This methodology gave insight into the way in which external product features affect the human perception of quality. Ti he research found [hat as the classification involved more product properties and became more complex, the error of human classification increased. Leemauis etal. (1 99) investigated the defect segmentation of Golden Delicious' apples using machine vision. The proposed algorithm was found to be effective in detecting various defects such as bruises, russet, scab, fungi or .vQujnds, In similar studies Yang (1996) assessed the feasibility of using computer Vision for the identification of apple stems and calyxes which required automatic grading and coring. Back propagation nei ral networks were used to classify each pat di as sternfcalyx or patch- like blemish. Earlier studies proposed the use of a `ff ding' algorithm to segment patch-like defects (russet patch, h raise, and also stalk or calyx area) (Yang 1994)_ It was found that this 06 method of feature identification is applicable to other types of produce with uniform skin colour. This technique was improved by Yang and Merchant (1995), who applied a `snake' algorithm to closely surround the defects. To discriminate russet in `Golden Delicious' apples a global approach was used and the mean hue on the apples was computed (Heinemann et at. 1995). A discriminant function sorted the apple as accepted or rejected, The accuracy reached 82i%. which is poor compared with European standards (Fieineman.n et al. 1995). Other studies involving Golden Delicious' apples were performed for the purpose of classification into yellow or green groups using the 1181 (hue, saturation, intensity) colour system method (Tao at al, 1995x). The results show that an accuracy of over 90% was achieved for the 120 samples tasted. Steinmetz et al. (1999) investigated sensor fusion for the purpose of sugar content prediction is apples by combining image analysis and near-infrared spectraphotometic sensors, The repeatability of the classification technique was improved when the two sensors were combined giving a value of 78% for the 72 test samples. An online system with the use of a robotic device (Molto ct aI. 1997) resulted in a running time of 3.5 s per fruit for the technique. Ruiz et al. (1996) studied three image analysis methods to solve the problem of long stems attached to mechanically harvested oranges. The- techniques include colour segmentation based on linear discriminant analysis, contour curvature analysis and a thinning process which involves iterating until the stem becomes a skeleton. It was found that those techniques were able to determine_ the presence or absence of a stem with certainty. A study by Kondo (1995) investigated the quality evaluation, Le, by the correlation of appearance with sweetness, of l okan oranges using image processing and reported that the method could effectively predict the sweetness of the oranges. Nagata et aI. (1997) investigated the use of computer vision to sort fresh strawberries, based on size and shape. In their study they developed a sorting system with an accuracy of 94-98% into three grades based on shape and five grades on size. However, another automatic strawberry sorting system was developed by Bata et al. (2000), had average shape and size accuracies of 98 and 100%, respectively. Fruit Shape is one of the most important quality parameters for evaluation by customer's preference.. Addition-ally isshaped fruits are generally rejected according to sorting standards of fruit. This Rashidi et al. (2007) was 67 carried out to determine quantitative classiii ation of algorithm for fruit shape in kiwifruit (Actinidia deliciosa) while Riyedi et al. (2007) classified papaya size according to combinations of the area, mean deter $nx p mtej features and they studied the uniqueness of the exacted features. The proposed technique showed the ability to perform Papaya size classification with more than 4% aceura ey. In another study by Alfantni et. al, (2008) developed an automated grading system for oil palm bunches using the RGB colour model to ditingcish between the three different categories of oil palm fruit bunches. Jn their study, result showed that the ripeness of fruit hunch could be differentiated between different categories of fruit bunches based on R B intensity. Pearson and Slaughter (1996) developed a ma ins vision system for the detection of early spilt lesion an the Dull of pistachio nuts and reported that the developed system classified early split nuts with 100" o a-uc cess and normal nuts with 99% acctraey. In other research a multi-structure neural network (MSNN) classifjier was proposed and applied to classify four varieties (classes) of pistachio nuts by Ohazanfari et al, (1996), An automated machine vision system was developed to identify i.ad remove pistachio nuts with closed shells from processing streams (Pearson md Toyofuku 2000). The Myst=m included a novel material handling system to feed nut, to line scan cameras without tumbling. The classifoation accuracy of this machine vision system for separating open shell from closed shell nuts was approximately 95%, similar to mechanical devices. Nielsen et al. (1998) developed a teebnique to correlate the attributes of size, colour, shape and abnormalities, obtained from tomato images, with the inner quality of the tomato samples. They applied fuzzy sets into their study. Recently, chaos theory was introduced into this area CMorimoto et al. 20D0). In this study t mats fruit shape was quantitatively evaluated using an attractar, fractal dimension and neural networks. The results showed that a combination of these three elements offers more reliable and more sophisticated classificatioc. Computer vision has also been used in the assessment of tomato seedling quality as a classification technique to ensure only good quality seedlings were transplanted (Ling and Rzhits1y 1996). The classification process adopted an adaptive tlaresholding technique, the Oust method, G8 As consumer awareness and sophistication increases the importance of objective measurement of quality is ever increasing. In a study by Miller and Delwiche (1989) the maturity of market peaches was evaluated by colotu analysis and the a mbined image processing with a finite element model to determine the firmness of pears was developed by T}ewulf et al. (1999). he application of computer vision technology to detect pear bruising was studied by Zhang and Deng (199). Results from the experiments confirmed that difereut bruised areas can be precisely detected with most relative errors controlled to within 10%. However, all above discussed quality evaluating and other commercial sorting machines that are currently available for commodities (cherries, nuts, rice etc.) are not capable of handling and sorting pomegranate arils for different reasons, thus making it necessary to build specific equipment. This work Blasco et aI. (2009) describes the development of a eornputer vision-based machine to inspect the raw material coming from the extiactiou process and classifies it in four categories. The machine is capable of detecting and removing unwanted material and sorting the arils by colour, 2.4.3.2. Assessment of ryCIab1es Computer- vision has been shown to be a viable approach to inspection and grading of vegetables (Shearer and Payne 1990). Heinemann at al. (1994) assessed the quality features of the common white Agaricus bisporus mushroom using image analysis in order to inspect and grade the mushrooms by an automated system. Tb it study showed that the disagreement between human inspectors ranged from 14 to 36% while by the vision system it ranged from 8 to 56%. Reed et al. (1995) found that computer vision could be combined with harvester technology to select and pick mushrooms based on size. Computer vision has also been applied to objective measurement of the developmental stage of mushrooms (L n 1996). This study found that cap opening of mushrooms correlated the best with the stage of development except for tightly closed mushrooms. Other research described the development of computer vision techniques for the detection, selection, and tracking of mushrooms prior to harvest (Williams and Heinemann 1998). From the spectral analyses on the colour of different mushroom diseases Vithanyo and Tillett (1998) concluded that the colour of the developed, seneSCent.mushroom differs from any browning caused by diseases allowing earlier detection of infected specimens. 69 Similar research developed a method, involving a series of complex Cnlour operations, to distinguish the diseased regions of mushrooms from naturally senescing mushrooms (Vizhanyo and Felfoldi 20O)).. Intensity normalisation and image transformation techniques were applied in order to anharce colour differences in true-colour images of diseased mushrooms. The method identified all of the diseased spots as diseased' and none of the healthy, senescent mushroom parts were detected as `diseased'. Potatoes have many possible shapes which need to be graded for sale into uniform classes for different markets. This created difficulties for shape separation. A Fourier analysis based shape separation method for grading of potatoes using machine vision for automated inspection was developed by Tao et al. (l995b), A shape separator based on harmonics of the transform was defined, Its accuracy of separation was 89% for 120 potato samples, in agreement with manual grading. Earlier. Lefebvre et al. (1993) studied the use of computer vision for locating the position of pulp extraction automatically for the purpose of furtherer analysis on the extracted sample. An image acquisition system was also constructer! for mounting on a sweetpotato harvester for the purpose of yield and grade monitoring (Wooten et al. 2000) It was found that culls were differentiated from saleable sweet potatoes with classification rates as thigh as 84%. Chilli is a variety grown exteasiveiy consumed by almost all the population. It has a high processing demand and proper sorting is required before filling or canning. A sorter that, Hahn (2002) classifies chilli by three different width sizes, by means of a photodiode scanner, was built they reported that the accuracy on the necrosis detection and width classification was of 96.3 and 87%, respectively. Color and size are the most important features for accurate classification and sorting of itrvs ( hojastebnazhand et al. 2010) and studies of the reflectance properties of citrus fruit (Gaffney 1973) can be considered as the bases for subsequent work on image analysis for citrus inspection. These studies determined the visible and near-infrared wavelengths at which meter contrast between the peel and major defects can be achieved (Motto and Blasco 2008). A simple technique was proposed by Cerruto et al. (1996), which segment blemishes in oranges using histograms of the three components of the pixel in HIS (Hue, Saturation, Intensity) color space. To estimate the maturity of citrus, Ying et al. (2004) used a dynamic threshold in the blue component to segment between fruit and background. They then used neural networks to distinguish between mature and immature fruit. However, in 70 another study, Khojastehnazhand et al (2010) developed an image processing technique for estimating citrus fruits physical attributes including diameters, volume, mass and surface area using image processing technique (I hojastehnazluand et al. 2O0) and they reported that this image processing procedure can be readily applied to other axi-symmetric agricultural products such as eggs, pearl, pepper, carrot, limes and onions, The main advantage of these techniques is their robustness when dealing with changes of color. Some other earlier studies of computer vision associated with vegetable grading and inspection include colour and defect sorting of bell peppers (Shearer and Payne 1990). Morrow et al, (1990) presented the techniques of vision inspection of mmshrooms, apples and potatoes for size, shape and colour. The use of computer vision for the location of stem/root joint in carrot has also been assessed F3atchelor and Searcy 1989). Feature extraction and pattern recognition techriiques were developed by Howarth and Searcy (1992) to characterize and classify carrots for forking, surface defects, curvature and brokenness. The rate of ndisclassification was reported to be below 15% for the 250 samples exanmin . 2.4.3.3. For grain elan ific.atiun and quality evaluation Wheat Grain quality attributes are very impDrant for all users and especially the milling and baking industries. Computer vision has been used in grain quality inspection for niauy years. An early study by yas et al. (1989) used machine vision to identify different varieties of wheat and to discriminate wheat from non-wheat components. In later research Zayas et al. (1996) found that wheat classification methods could be improved by combining morphoznetry (computer vision analysis) and hardness analysis. Hard and soft recognition rates of 94% were achieved for the seventeen varieties examined. Twenty-three morphological features ixere Used for the discrf ni ant analysis of different cereal grains using machine vision (I ajumdar et al, 1997). Classification accuracies of 98, 91, 97,100 and 91% were recorded for CWRS .(Canada Western Red Spring) wheat CWAD (Canada Western Ambcr Durum) wheat, barley, oats and rye, respectively. The relationship between colour and texture features of wheat samples to scab infection rate was studied using a neural netwo& method (Run et al. 1997). It was found that the infection rates estimated by the system followed the actual ones with a correlation coeffkient of 0.97 with human panel 71 assessment and maximum and mean absoIute errors of 5 and o, respectively. In this study machine vision-neural network based technique proved superior to the human panel. [mage analysis has also been used to classify dockage components for CWRS (Canada Western Red Spring) wheat and other cereals (Nair et aI. 1997). Morphology, colour and morphoIogylcoIour models were evaluated for classifying the dockage components. Mean accuracies of 89 and 96% for the morphology model and 71 and 75% for the colour model were achieved when tested on the test and training data sets, respeltively, Cmrn In order to preserve corn quality it is important to obtain physical properties and assess mechanical damage so as to design optimum handling and storage equipment. Measurements of kernel length, width and projected area independent of kernel orientation have been performed using machine vision (Ni et al. 1997a). While, Steenbok and Precetti (2000) performed a study to evaluate the concept of two-dimensional image analysis for cIassification of maize kernels according to size category and they reported that the ciassification accuracy of both machine vision and screen systems was above 96% for round- hole analysis. However, sizing accuracy for flatness was less than 80%. Ng of al. (1997) developed a machine vision algorithm for corn kernel mechanical and mould damage measurement, which demonstrated a standard deviation less than 5% of the mean value. They found that this method was more consistent than other methods available, The automatic inspection of corn kernels was also. performed by Ni at al, (199Th) using machine vision and they showed classification rates of 91 and 94% for whole and broken kernels respectively for whole and brokein kernel identification daring their on-line tests study. The whiteness of corn has been measured by an on-line computer vision approach by Liu and Paulsen (1997) and they were this technique easy to perform with a speed of 3 kernels/sec. In other studies Xic and Paulsen (1997) used machine vision to detect and quantify tetrazolium staining in corn kernels. The tetrazolium- machine vision algorithm was used to predict heat damage in corn due to drying air temperature and initial moisture content. Conclusively it could be say that the machine vision system is playing a versatile role in quality evaluation earn so for. Rice and lentils 72 As rice is one of the leading food crops of the world its quality evaluation is of importance to ensure it remains appealing to consumers. Liu et aI. (1997) developed a digital image analysis method for measuring the degree of milling of rice and they obtained the coefficient of determination of R2 =0.98 for the 60 samples. Wan at al. (2000) employed three online classification methods for rice quality inspection: range selection, neural network and hybrid algorithms. The highest recorded online classifiication accuracy was around 91% at a rate of over 1200 kernels/min. The range s&tectinn method achieved this accuracy but required time-consuming and complicated adjustment. In another study, milled rice from a laboratory mill and a commercial-scale mill was evaluated for head rice yield and percentage whole kernels, using a shaker table and a machine=vision system called the Grain Check (Lloyd et a1. 2000). however, a flat bed scanner was used in machine vision system developed by Shahin and Symons (2001) for colour grading of lentils. In their study, they scanned and analyzed different grades of large green lentils over a two-crop season period and developed an online classification systen1 based on neural classifier. 24,3.4. Processed food products Visual features such as colour and size indicate the quality of many prepared consumer foods. Sun (2000) investigated this in research on pizza in which pizza topping percentage and distribution were extracted from pizza images. Then it was foxmd that the, new region-based segmentation technique could effectively group pixels of the wine topping together and the topping exposure percentage can be easily determined with accuracy 90%. To avoid the miss-guideness of quality assessment by visual based human perception, computer vision has been widely used in the assessment of confectionary products so far, Davidson et al. (2001) measured the physical features of chocolate chip biscuits, including size, shape baked dough colour, and fraction of tap surface area that was chocolate chip using image analysis. Four fuzzy models were developed to predict consumer ratings based on three of the features. A prototype-automated system for visual inspection of muffins was developed by Abdullah et al. (2000) and they reported that it was able to correctly classify 96 of pre graded and 79% of unwed muffins with an accuracy of greater than 88 o. Now, the machine vision system has also been used in the assessment of quality of crumb grain in bread and cake products [ apirstein 1995). 73 Cheese The evaluatioi of the functional properties of cheese is assessed o ensure the necessary quality is achieved, especially for specialized applications such as consumer food tappings or ingredients. Wang and Sun (2001) developed a computer vision method to evaluate the melting and browning of cheese, This novel non-contact method was employed to analyze the characteristics of cheddar and mozzarella cheeses during cooking and the results showed that the methed provided an objective and easy approach for analyzing cheese functional properties (Wang and Sun 2002a, 2002b). Ni and Guoasekaran (199) developed as image- processing algorithm to recognize individual cheese shred and automatically measure the shred length. It was fond that the algorithm recognized shreds well, even when they were overlapping. It was also reported that the shred length measurement errors were as low as 0.2% with a high of 10% in the worst case. Noodles The important of noodles in the Asian diet is very significant, as currently they account for 30-40% of most countries' wheat consumption (Miskel1 1993). The majorityof Asian noodle manufacturers noodle products to be derived from white seed-coated wheat as compared to red seed-coat material. Initial research had used image analysis to characterize the white wheat products (Hatcher and Symons 2000), As noodles appearance and color are the consumer's initial quality parameters, there is a preference for rnam ifazturers to use high quality, low yield patent flours and free from various contaminants of the, flour (i.e. bran and germ), as they result in brighter noodles. The standard methodology to assess bran contamination has been ash. content, the use of auto fuoresc ce by different wheat seed tissue (Muncie at aI. 1979). Hatcher et al. (1999) demonstrated that the image analysis could be effectively used to quantify, measure, and discriminate varietal differences in white seed coated wheat varieties in preparation of yellow alkaline noodles prepared from high-quality patent flour. Potato chips and Freiwh frf r The images of commercial potato chips were evaluated for various tolnur (Brosnan and Sun 2004) and textural features to characterize and classify the appearance edreschi et al. 2004) and to model the quality Preferences of a group of consumers. . 74 The most frequently used color model is the RUB model, in which each sensor captures the intensity of the Iight iii the red (R), green (G) and blue (B) spectra respectively, for food quality evaluation (Du and Sun 2004a). However, the features. derived from the image texture contained better information than colour features to discriminate both the quality categories of chips and consumers preferences. Table 2.8. Summary of machine vision applications for food and. agricultural produce. Produnt products Application Accuracy Efficiency e Fruits Apple Defects detection - -- Paulus and Schrevens (1999) Classification 78% Steinmetz et al. (1999) Grading 95% Yang (1994) Oranges Classification 93% Ruiz et al. (1996) Quality Evaluation 87% Kondo et at (1995) Strawberries sorting 94-93 agate et al. (1997) 98-1009% Kato et al. (2000) Oil Palm Sorting — Alfantni (2008) Papayas Classification 94% Riyadi at al. (2007) Nuts Classification 95% Pearson and Toyoful u (2000) Pomegranate Sorting (Aril) 90° Blasco et al. (2009) Vegetables Potatoes Classification 84° Wooten et al. (2000) Chilli Classi icatioa 87% (2002) & 96.3 Lemon Classification and --- Lino et al. (2008) sorting eats Cerra Size 73-90% Liu and Paulsen (1997) Vihiteness ie and Paulsen (1997) Whole and broken 91-94% Ni et al, (1997a, b) kernel Grading 30-96% Steenhoek and Precetti (2000) Wheat, barley, Grading 9111/0 Wan et al. (2000) oats, rye Classification 98, 97, Majumdar et al. (1997) l0D and 91 75 Fedreschi et al. (2006) recently designed and implemented a computer vision system to measure representatively and precisely the color of highly heterogeneous food materials, such as potato drips. Leon et al. (2006) developed a computation Dolor conversion procedure that allows the -obtaining of digital images from the ROB images by testing five models: linear, quadratic, gamma, direct and neural network. After the evaluation of the performance of the models, the neural network model was found to perf rrn the best, In another study, a stepwise logistic regression model was developed by Mendoza et al, (2007) and they reported that that it was able to a cplaixi 86.2% of the preferences variability when classified into acceptable and non-acceptable chips. Meat and meatprodiwts Visually discernible characteristics are routinely used in the quality assessment of meat. McDonald and Chen (1990) pioneered early work in the area of image based beef grading. Based on reflectance characteristics, they discriminated between fat and lean in the Long simus muscle and generated binary muscle inages. In a more recent study Gerrard et al. (1996) examined the degrees of marbling and colour is 50 steaks and they predicted the lean colour (R2=0.6) and marbling scores (R .84). Image texture analysis has also been used in the assessment of beef tenderness (Li et al. 1997). Statistic regression and neural network were performed to compare the image features and sensory scores for beef tenderness and it was found that the textore features considerably contributed to the beef tenderness. In another study by Lu et al. (2000, 1997). evaluation of pork -quality has also been investigated and recommended neural network modeling as an effective tool for evaluating fresh pork colour. Furthermore, Gray-scale intensity, Fourier power spectrum, and fractal analyses were used as a basis for separating tumorous, bruised and skin torn chicken carcasses from normal carcasses (Park at al. 1996). A neural network classifier used performed with 91% accuracy for the required separation based on spectral images scannedat both 542 and 700 nit wavelengths. In a further study Park and Chen (2001) found that a linear discriminant model was able to identify unwholesome chicken carcassca with classification accuracy of 95,6% while a quadratic model (97% accuracy) was better to identify whQlesorne carcasses. Daley 7G A t. o...... ,.. j ., %. • r et al. (1994) analyzed chicken carcasses for systemic defects using glo based on a neural network classifier. However, the correct assessment of meat quaIity, to fulfill the consumer's needs, is crucial .element within the meat industry. Although there are several factors that affect the perception of taste, tenderness is eoz sidered the most important characteristic. Corte2a et ai. (2006) presented, a feature selection procedure, based on a Sensitivity Analysis, is combined with a Support Vector Machine, in order to predict lamb meat tenderness. This real-world problem is defined in terms of two difficult regression tasks, by modeling objective (e.g. Wamer-Bratzler Shear force) and subjective (e.g. human taste panel) measurements. In both cases, the proposed solution is competitive when compared with other neural (e.g. Multilayer Perceptron) and Multiple Regression approaches. 2.4.3.5. Detectlon and local3zat4an of fruits on a tree/plant To detect the fruit on a tree more efficiently and to increase harvesting operations, many fruit harvesting systems have been developed in the last decade. Among them is robotic fruit harvesting. The vision based fruit harvesting system basically depends on the contribution of different features in the image. The 4 basic features which characterize the fruit are intensity, color, edge, and orientatiorx, Efficient locating of the fruit on the tree is one of the major requirements and a tough task for the fruit harvesting system. Patel et al. ( O11 b) have developed an image processing algorithm based on multiple features to detect fruits on trees. They designed an image processing algorithm to calculate different weights for features like intensity, color, orientation, and edge of the input test image. Based on weight of these features they approximated the locations of the fruit within an image and achieved detection efficiency up to 90% for different fruit images an the tree captured from different positions. However, image segmentation based on color difference between mature fruits and backgrounds under natural illumination condition is a difficult task. So. to recognize the tomato fruits, stems, leaves, and stem-supporting pole, image processing was carried out in three colors spaces of ROB (red, green, and blue), HI (hue, saturation, and intensity) and YCbCr (Luma and (hrama signals) from a CCD camera by Monavar at a1. (2011). They compared the processed images in three color spaces in order to identify ripe tomatoes with more than 50% redness and concluded that the ROB and HSI spaces may be 77 implemented to realize dynamic threshold segmentation of color of fruit images in which color of objects and background show a noticeable differ nce. It makes fruit image segmentation easy and changes an original 3D color image into a iD gray-level image segmentation, which can be used to remove unvalued data of images, thus increasing speed and efficiency of sorting aigorithms. To convert colors from I GAB to HSI, the equations (Abbasgholipour et al. 2011) generally used in image processing arc: 4 BAG (4) 360— B?G =l- + t {R G B) (5) IJ(R+G +B) (6) where B=Cas ((R— )'+(R— )XR—E)]l~z namulai et al. (2004) investigated machine vision utilizing color vision as a means to identifying citrus fruits and to estimate yield information. They converted the KGB image into HIS and developed a computer vision algorithm to enhance and extract the information from the images. They estimated threshold segmentation of images from pixel distribution in HSI color plane. In another study, Jianjun et al (2006) have presented an automatic segmentation method of tomato image that comprises of three main steps. Firstly, Red and Green campnnent image are extracted from kGB color image, and Green component subtracted from Red component gives KG of chromatic, aberration gray-level image. (ha - level value between objects and background has obviously difference in KG image. By the feature, Ostu's threshold method is applied to do adaptive KG image segmentation. And then, marker-controlled watershed segmentation based on morphological gray le reconstruction is applied into Red component image to search boundary of connected or overlapped tomatoes. Finally, intersection operation is done by operation results of above two steps to get binary image of final segmentation. They have found it very robust and effective to segment multi-tomatoes image of various growth states under the natural illumination condition. Furthermore. Vanhentsn et at (2 02) have presented a concept of the modular cucumber harvesting robot. They tested the harvesting robot in a greenhouse with a success rate of 8O°A which confirmed its ability to pick cucumbers without human interference. The computer vision system is able to detect more than 95% of the cucumbers in a greenhouse, On average the robot treaded 45 s to pick one cucumber. Future research focuses on hardware and software solutions to improve the picking speed and accuracy of the eye-hand co-ordination of the robot. - Baeten at al (2007) have described the ' anstruction and functionality of an Autonomous Fruit Picking Machine (AFPM) for robotic apple harvesting and suggested that the future work on will need to focus on improving to optimizing the image processing in spite of they were Found first results of the AFPM very promising. 2.4.3.6. Object measurements and classification Size and Shape Size (length, width, area, and perimeter) and shape are 2 important geometric measurements of food products in the food industry and they carry the direct information that can be used for quality evaluation and inspection. Uniform sites and shapes of fruits for export are required in order to ease handling and transportation, and also to help satisfy consumer preferences. So, besides color and texture, size and shape can also be extracted as primary types of object measurements acquired from images (Du and Sun 2004a). However, size measurements of objects in digital images are restricted to being one-dimensional (1-D) and two-dimensional (2-]D). The three-dimensional (3-D) information .regarding the objects is lost during image acquisition, unless special techniques such as structural lighting are used (Saxes 1994). On the other hand, shape is an important geometric measurement of food products and plays an important part in purchase decisions by customers eemans and Destain 204). Evaluation of product acceptance by customers using machine vision learning techniques (Du and Sun 2004a,b, 2006b) and discrimination of the products with different characteristics (Zion et al. 1999, 2000) are typical applications of shape. 79 W_ Jarimopas and Jaisin (2008) have developed a machine vision system to sort sweet tamarinds according to shape (straight, slightly curved, and curved), 'size (small, medium, and large), and defects. Using software, they characterized shape by shape indexes, size by Image length, and defects by grey scale profile and reported that the control factors (belt conveyor sperm, pod spacing, and o ntation) did not significantly affect the giality parameters. Vaysse at al. (2005) introduced a new imago analysis tool based on the optimal positioning of circles an the gradient magnitude computed for the Iuuninauce of the image in the quality assessment of frulis placed in bulk shipping bins. According to authors, once the fruits are detected, it is possible to measure such criteria as size or color. SIameI et al. (2007) classified papayas according to combinations of the 3 features to study the uniqueness of the extracted features. The proposed technique slowed the ability to perfDrm papaya sin classification with rpare than 94% accuracy. An on-line method based on imaging teo h.niques was developed by Labba.fi et al. (2007) for measuring mean bubble size and bubble size distribution in food foams manufactured in a continuous foaming device. They validated this method on 4 kinds of aerated food emulsions (ice crearn, whipped cream, aerated fresh cheese, and a foamed sauce), corresponding to a large range of overrun (30— ISO%). l-cic at al. (1998) developed an accurate method for measuring of the length, width, and shape of c mibers, They determined the local width along the midline of a cucumber to obtain a condensed description by which length and width can be extracted. Sapirstein and Kohler (1999) investigated the effect of sampling on the precision and accuracy of digital image analysis of different commercial sample grades of Canada Wsterzi Red Spring (CWRS) wheat. Kernel perimeter, length, width, and area measurements were used to determine mean and dispersion statistics of grades of CWRS samples. The length and width of an object can also be used to measure the size of are object which is scessary to locate the major axis of the object and to measure its relative length and width. Monnin (1994) designed .a system for a multiple-Iane cracker baking process. The system used diameter, area, and thickness to measure the crackers, Vandevooren et al. (1992) statistically analyzed the measurements of length, width, and a range of more or 1e s complex shape descriptors to identify mushroom cultivars. However, the calculation of Iength and width of irregularly shaped food products is more complex than that of area and perimeter. Nevertheless, mcasurcrcnts most commonly used 80 are Feret's diameter (Fig. 9), the major axis, and the minor axis (Zheng et a], 7006b) developed by researchers. Feret's diameter is defined as the difference between the largest and smallest of the coordinates of an object at different rotations. x Fig. 2.9. Detcrrnination of Feret's Diameter The measure and minor axes can also be defined as those in an ellipse which is fit to the object Using ellipse-fitting methods (Mulchrone and Cloudlrnry 2004; Zheng et al, 2006). Cober et al. (1997) measured the seed size (cross-sectional area) in 2 erie ations to estimate heritability of seed size in soybean. Orientation is a drawback to length and width measurements that needs constantly to be updated Recently, actual projected area and surface area were measured using the image processing technique (Soltani -et al. 2.010) and it was reported that the estimated volume and prolectcd area were not significantly different from the volume determined using the water displacement rnethod(P > 0.05) and the projected area measured by the IF technique (P> 0.05), respectively. However, the estimated surface area was significantly different from the measured surface area by the IP method. Weight and volume An image-processing based technique was developed by Omid et al. (2010) to measure volume and mass of citrus fruits such as lemons, limes, oranges, and tangerines. They designed an efficient algorithm and implemented it in visual basic ( ) language. The product volume was calculated by dividing the fruit image into a number of elementary elliptical frustums and showed good agreement with the actual volumes determined by the water displac nient method. The coefficient of determination (R2) for lemon, lime, orange, and tangerine were 0.962, 0.970, 0.985, and 0.959. respectively. They concluded that the 81 results indicated that citrus fruit size has no effect on the accuracy of computed volume and proposed a simple procedure based on the computed volume of an assumed ellipsoidal shape for estimating the mass of citrus fruits. Spreer and Muller (2010) proposed an equation in order to calculate the mass of mangos based on 3 geometric dimensions (length, width, and thickness). It showed a high level of explanatory power of R2 — 0.97. Based on measurements taken from the digital photos, the equation was then used to determine the estimated mass based on the 3 dimensions. Estimated and measured values were. plotted against each other. A high correlation (R2 = 0.96) was found between measured and calculated mass. Authors showed that the 3 most characteristic dimensions, namely L, VUmax, and Tmax, are sufficient to describe the size and weight characteristics of a mango. color Color is elementary iaforrnation that is stored in pixels to constitute a digital image. Color is therefore rated as one of the most important object measurements for image understanding and object description. According to b-ichrornatic theory, that color can be discriminated by the combination of 3 elementary color components (Young l&)2; MacAdam 1970), 3 digital values are assigned to every pixel of a color image. Color features of an object can be extracted by examining every pixel within the object boundaries. Color has proven successful for the objective measurement of many types of food products, with applications ranging from fruit, grain, and meat to vegetables (Du and Sun 2004a). Color has been successfully used in measuring fruits. Miller and Delwiche (19S9) developed a color vision system to inspect and gradefresh-market peaches, Diffused lighting and normalized luminance were used to reduce the red, green, and blue inputs to two- diniensienal chromaticity coordinates. Peach color was compared with standard peach maturity coi!rs for classification. Leemans et a1. (1998) used color machine vision for defect segmentation on Golden Delicious apples. A colour model was used as a standard for comparison with sample images. The color of tomatoes was used for estimating the maturity stage using image analysis (Choi et al. 1995; Jalhns at al, 2001). Color is used extensively for the measurement of grain. Casad r et al. (1992) developed a trainable algorithm on a color machine vision system for the inspection of 82 swybean seed quality. They used color chromaticity coordinates and seed sphericity for seed classification. White and Sellers (1994) developed a high-speed color inspection system to detect foreign materials on a peanut conveyor belt in real time and reported that the system is trainable to recognize and differentiate unique color signatures, or fingerprints of many types of foreign materials and food products, A RGB color feature-based multivariate decision model to discriminate between asymptomatic and symptomatic soybean seeds for inspection and grading, which comprises 6 color features including averages, rninimwns, and variances for ROB pixel values, was developed by Ahmad (1999), In that study, 88% overaIl classification accuracy was achieved by using a linear discriminant function for asymptomatic and symptomatic seeds with highest probability of occurrence. In addition to this, an automatic system was developed by Ruan et al. (2001) to determine the weight percentage of scabby wheat kernels, lased on color features of scabby kern.eIs captured by machine vision, So, color can be implemented for the automated inspection and grading of fruits and cereals with good accuracy and efficiency in real time. Furthermore, Daley and Thompson (1992) automated the inspection and grading of meat products to help meet the high output and improve objectivity in the industry. Park and Chen (1994) used 6 optical filters at different wavelengths to obtain multisp coral images for poultry carcass inspection and they investigated the potential of using colour image processing to detect defects such as bruises, skin tears and systemic diseases. Lu et al. (2000) extracted color image features such as mean and standard deviation of red, green, and blue bands from segmented images for evaluating fresh pork. In spite of the above discussions, color features can also be used to quantify the distribution of ingredients for quality classification using computer vision. Chan et al. (2007) designed a law-cost visual-based classification system, consisting of a- color webcam interfaced with the computer to understand the use of color in machine vision and to enhance color processing in PC-based vision systems. The entire system operates on a Ma lab-driven program that analyzes the color of the objects being inspected. For this system, they inc1uded the process of dithering, calculation of the mean values, and histograms, and the final result as theory of operation. One of the potential applications is to segregate the ripe and unripe fruits, This system is fundamentally capable of identifying 4 types of tropical fruits .(mango, banana, papaya and tomato). A computer vision system (CATS) consisting of a standard lighted box, a camera for image acquisition, and imago processing soft=re was developed by Laurent et al. (2010). They analyzed red bean images based on RGB and luminance histogram determinations, and achieved histogram feature definitions and determinations. On the basis of a left shift of histogram with variation in color of the beans they proved browning of red beans during storage. Ebrahimi et al. (2011) used machiae vision to analyze the greening area of potatoes based on ROB (Red-Green-Blue) and HIS (Hue-Saturation-Intensity) spaces. The difference been red and men ompenents oft B space for green parts of potatoes was lower than that of other parts. They proposed a machine vision-lased algorithm in ROB and HIS spaces, Hashim et al. (2010) demonstrated the potential of application of 1 $ imaging to detect chilling injury symptoms in banana. They used zero- and first-order kinetics models to describe the color change kinetics and observed that r fitted well to the zero-order model, while g and hue followed the first-order, All the color spaces except b showed high ,orrelation with visual assessment. rage Lure analysis Image texture analysis is another active research topic in computer vision and image ocessing. Texture effectively describes the properties of elements constituting an object's Grface, thus texture measurements are believed to contain substantial information for the ittem recognition of objects (Amadasun and King 1989). Although texture can be roughly fined as the combination of some innate image properties, including fineness, coarseness, aoothuess, granulation, randomness, lineation, and so on, etc., a strictly scientific definition r texture has still not been determined [Haralick 1979). Accordingly, there is no ideal ethod for measuring texture. Nevertheless, a great number of methods have been :ve1oped, and these are characterized into statistical, structival, transform-based, and odel-based methods (Zheng et al. O0 a). Among the texture analysis methods for food tality evaluation, most approaches are statistical including the pixel-value run length ethod (Li et al, 1999 Majumdar and ]ayes 2000a) and the co-occurrence matrix method McCauley et al. 1994;'Park and Chen 1996; Sh ranita et al. 199$). However, several texture 84 description methods are based on Fourier spectrum, wavelet transform, and fracta~ dimension (Li et al. 2001; Quevedo et al. 2002; Whittaker et al. 1992). - Statistical methods compute different properties and provide textural characteristics such as iioothness, coarseness, and graininess. They are suitable when texture primitive sizes ire comparable with pixel sizes. McCauley et at. (1994) used co-occurrence image texture for both prediction and classification of intramuscular fat in beefIron' uItrisouic images of both live beef wimals and slaugbtered caresses. According to Park and Chen (1996), it is feasible for differentiating abnormal from normal poultry carcasses at a wavelength of 542 nm with textural feature analysis of multispectral images containing visible/near-infrared wavelengths based on co-occurrence matrices. Shirariita et al. (199S) described a method of determining meat quality using the concepts of marbling score, a measurement of the distribution density of fat in the rib-eye region, and texture analysis. They evai ated the grade of unevaluated images by comparing the texture feature vector of an image with the standard texture feature vectors and reported that the system is effective. Furthermore, to predict cooked-beef tenderness from fresh-beef image characteristics, Li at al. (1999) xtmctd beef muscle image texture features by using pixel value run length and pixel value spatial dependence methods. In their study they found that texture features were useful indicators of beef tenderness. Another convenient way to determine the texture of an object is to examine its spatial frequency content. Whittaker at al, (1992) used Fourier transform to anafyze an image with texture features, extracted from ultrasonic images for the prediction of a beef marbling scare. According to research studies, wavelet textural analysis can be viewed as a combination of spectral and statistical methods. For instance, Li et al. (2001) used a waveIet-based decomposition method to extract texture features of fresh-beef images, which were used to classify steaks into tough and tender groups in terms of cooked beef tenderness. Fractal technology can also be used for determining texture features by analyzing surface intesity using the fractal dimension, namely, the fractal texture. Quevedo et al. (2002) used fractional Brownian motion, box counting, and fractal dimension (El)) estimation from the frequency domain to numerically describe the surfaces of foods and the miorostructuue of potato cells for texture analysis. as 2.4.4. Advantages and disadvantages Machine vision system is seen as an easy and quick way to acquire data that would be otherwise difficult to obtain manually (Lefebvre at al. 1993). Since, the capabilities of digital image analysis technology to generate precise descriptive data on pictorial information have contributed to its more widespread and increased use (Sapirstein 1995). Quality control in combination with the increasing automation in all fields of production has led to the increased demand for automatic and objective evaluation of different products. SistIer (1991) confirm that computer vision meets these criteria mmd states that the technique provides a quick and objective means fir measuring visual features of products. In agreement it found that a computer vision system with an automatic handling mechanism could perform inspections objectively and reduce tedious human involvement (Morrow et al. 1990). In other study, Gerrard et a1. (1996) also recugaized that machine image technology provides a rapid, alternative means for measuring quality consistently. Another henefit of machine vision systems is the non-destructive and undisturbing manner in which information could be attained (Zayas et ai. 1996). making inspection unique with the potential to assist humans involving visually intensive wort (Tao et at. 1995b). Tarbell and Reid (1991) noted that an attractive feature of a machine vision system is that it can be used to create a permanent record of any measurement at any point in time. Hence archived images can be recaLIed to look at attributes that were missed or previously not of interest. Human grader inspection and grading of produce is often a labour intensive, tedious, repetitive and subjective task (Park et al. 1996). In addition to its costs, this method is variable and decisions are not always consistent between inspectors or from day to day (Tao at al. 1995a; Heinemann et al. 1994). In contrast Lu et al. (2000) bad found computer vision techniques adoptable, consistent, efficient and cost effective for food products. Hence computer vision could be used widely in agricultural and horticulture to automate many labour intensive processes ( unasekaran 2001), Furthermore, unasekar n and Ding (1993) also bad agreed that machine vision technique's popularity in the food industries is grrwir~g constantly and they also pointed out that its development so for at a robust level and competitively priced sensing too (Yin and Pan.igraiii 1997). Moreover, all above advantages, there are more other advantages of machine vision system to different agriculture sectors, are summarized bellow in Table 2,8, An ambiguity of computer vision is that its results are as influenced by the quality of the captured images. Often due to the unstructured nature of typical a ,ricu1tural settings and biological variation of plants within them, object identification in these applications is considerably more difficult. Also if the research or operation in being conducted in dim or night conditions artificial lighting is needed. Some advantages and disadvantages of computer vision to different sectors of the agricultural and horticultural indumries have been sum mat d in Table 2.9. So, according to above findings, computer vision has been widely used for the sorting and grading of fruits and vegetable. It offers potential to automate manual grading practices and thus to startdai 1ize techniques and eliminate tedious inspection tasks. Karnali et al. (1998) reported that the automated inspection of produce using machino vision not only results in labour savings, but can also improve inspection objectivity. Table 2.9. Benefits and drawbacks of machine/computer vision system Benefits and drawbacks References Benefits Generation of precise descriptive data Sapirteiri (1995) Quick and objective Li et a1. (19}7a, b) Reducing tedious human involvement Ni et al. (199Th) Consistent, efficient and cost effective Lu et al. (2000) Automating many labour intensive process unasekaran (2001) Easy and quick, consistent Gerrard et al. (1996) Non-destructive and undisturbing Tao et al. (I 995x); Zavas er a1. (1996) Robust and competiti'iely priced sensing technique Gunasckaran and Ding (1993) Permanent record, allowing furtheranalysis later Tarbell and Reid (1991) Drawbacks Object identification being considerably more Shearer and Holmes (1990) . difficult in unstructured scents Artifieiai lighting needed for dim or dark conditions Stone and Kranzier (1992) Brasov and Sun (2004) 8T Chapter Materials and methods 3.1. Sample Collection Four orchards were selected based on flowering amount, size of trees (for easy harvest) and location (easily approachable) for the study of four varieties of mango (cv. Ckausa, Dashehari, Langra and Malda). Among these orchards, three (one for Chausa, orie for Dashehari and one for Langr i) orchards were from Utter Pradesh and one (for Makda) from Bihar states of India. Five trees, each from east, we,t, north and south, and centre of orchard were randomly selected. Ten fruits, two from each side and two from centre of tree were harvested manually at pre-r optimum and over maturity (when the majority of the mangoes were ripening on the tree) and all so collected 50 fruits were mixed thoroughly and 25 among them were picked randorily for experiment. Altogether 100 samples were collected from the orchards for experiment and they were brought to laboratory in a temperature controlled system (I 5°) and stored immediately in a cold room m.airitained at 20°C, 95% RH, till further experimentation. Flotation method (Jacobi et al. 1295) was used to confirm the maturity and subsequently graded based on maturity. floaters' were considered immature and "sinkers" as mature._ Only mature fruits were considered for experimentation at optimum and over maturity stage of harvesting. All the physical and biochemical analysis of the fresh samples were conducted in ambient conditions (28 ± 0.1°C and 70 ± 1% RH) within 24 h of harvesting. The samples were then stored at 20±2°C and 95+1% relative humidity (RH) and analysis were repeated at the interval of 3 day up to 15 days of storage. Two cultivars (i.e. Chausa and Dashehari) were selected for the non-destructive quality analysis of mango fruits on the basis of peel colour more suitable for the image processing. Computer vision system equipped with four cameras (monochrome, coIor, UV and N[R) oni-by-anti was used in this current research to check thr, potential of image processing technique in sorting/grading of mango fruits. For this purpose, 60 ripe fruits of both cultivars were procured randomly in the spring season of 2011 from the corresponding orchard and brought to Iaboratory. Using different cameras (one by one) ss and lighting regime, six images of each fruits were captured. A detailed description about computer vision system, its accessories, developed algorithms and steps involved in processing of images, etc. have been presented in the section 3.3. 3.2. Physt-ca! and chemical analysis 3.2.1. Determination of physical properties 3.2.1.1. Length, Breadth, Thickness Linear dimensions of selected samples, i.e. length, a; breadth, b and thiclne. s, c were measured using a digital vernier caliper (model no CD-ii" CSX, Mitutvyo-Japan) having least count of 0.01 mm. The arithmetic m•aan diameter (Da) and the sizes of mango in terms of geornetric mean diameter (Dg) were calculated using equation 3.1) and (3. 2), respectively (Mohsenin, 1975): a+btc (.1) 3 Dg =.abc)' (3.2) Where a — longest intercept; b longest intercept normal to a; c — longest intercept normal to a and 1,. 3.2.1.2. Sphericity and Aspect ratio The sphericity (p) and aspect ratio (Ra) were computed using equation given by Mo euiu (1978) and I aduako and Faborocde (1990) as below: Gccrnwtrici1 r ears diameter (3.3) Longest intercept The R.a(%) was obtained using following r'eIationship as recommended by Breadth ka(%) 100 x (3.4) Longest intercept 3.2.1.3. Mass, volume and speelfie gravity `lam mass of mango fruits were measured by a digital balance, with an u.Puracy 0.01 g and the volume of mango huts were measured using the water displacement technique (Mohsenin, 1936) based on Archimedes principle. Each mango was subim rg d in a 500 om3 water in eureka container and the volume of water displaced was measured 99 using gtduat d cylinder. Water temperature during measurements was kept at 25C° (Ra.shidi and Seyfi, 2007). After measuring the volume, the specific gravity of each fruit was computed using equation (5) as below: _ Mass in air (3. ) Volume of air Specific gravity is the density of a substance divided by the density of water. Since (at standard temperature and pressure) water has a density of 1 g/cm3, and since all of the units cancel, specific gravity is usually very close to the same value as density but without any units) (Anon 2012). 3.2.2. Determination of biochemical properties 3.2.2.1, Total soluble solids The total soluble solids were determined using digital ref actometer. Homogenous sample was prepared by blending the peeled mango flesh in blender. The sample was thoroughly mixed. The juice was filtered using a muslin cloth. New piece, of cloth was used for filtering each sample. A few drops were taken on prism of digital refractometer (model NR151 E TECH instruments) and direct reading was taken by reading the scale in meter as described in AOAC (1994) at room temperature. 3.2.2.2 .Titratable Acidity The titratable acidity of the sample was determined using method described in AOA .(1994). The whole mango pulps ground in mortar and pester and mixed properly. The 2g sample was weighed in conical flask amid the volume was made up as to lOOmI and filtered.. The 1OmI of the aliquot was titrated against 0. iN NaOH solution using phenolphthalein as indicator. The percentage acidity of every sample was calculated as: Titre N E 100 (.6) Total acid (%) vx Fix 1000 Where N = normality of alkali, V Volume made up, E — Equivalent of acid, v = Volume of sample taken for estimation and W , Wt. of sample taken 90 3+2.2.3. Total carotenoids Total carotenoids were estimated by method described by Alasalvar et al. (2005). Ton grams of representative sampI were blended in a household blender at high speed with 100 mL of acetone for 2 min in dim light to extract total carotene, The homogenate was filtered with suction through a Buchner funnel containing whatman filter paper No.! The resulting paste was extracted twice with acetone until the extract was colorless. The filtrates were combined, transferred to separatory funnel containing 50 mL of 4% aqueous Na CI and 100 niL of petroleum ether. The absorbance reading of the extract was taken at 450 um in a LTV- is spectrophotometer (make model), with petroleum ether layer as blank. After measuring absorbance, total carotenoids were computed using equation as below: Total carotenoids (rng I IOl g) — — — - (33) W( ) l000 Where O.D = Optical density at 450 nit. V = Volume made up (ml) ezid W = weight of the sample (g). 3.3 Nc n-destructive Analysis An imaging (computer vision) system (Fig. 3.1) fabricated in Division of Post harvest Technology at Indian Agricultural Research Institute, New Delhi-12 was used for nondestructive quality analysis of mangoes in this current research. The camera, which is placed at the top of the illumination chamber, acquires the image of the manually placed real mango fruits. The video signal is sent from the camera to the industrial computer, which is equipped with a frame grabber that captures the image. All images (visible dater} visible monochrome and UV) were stored in computer hardware memory in BMP except NIR stored in TIF form, The computer obtains the position of the centre of the detected objects and this information is used to process the eLenrnt. Thus, the imaging system that used for quality analysis of mango mainly consists of an illumination chamber, different image acquisition system (different lighting arrangements and different cameras setup), frame grtber, image processing and analysis software and computer hardware. 91 3.3.1. Computer Features 3.3.x.1. Hardware Industrial computer (PC type) with LED screen for desk mounting; desk type keyboard and mouse; TFT color Display of 15" XGA (1024 x 768); Pentium 4 1,7 Hz; 4GB of RAM; 4 serial parts RS-232; 4 TJSB ports; Hard Disk of 500 GB; 3 slot for PCI. 3.3.1.2. Sof twttr P Professjona, Microsoft Windows, with service pack 3 (5133) that is an important update and includes previously released security, performance, and stability updates for Windows XP and IMAQ Vision which is a library of L b JE (National Instrument product, Austin, TX) software (version 10.0) were used in image processing and analysis in this research work, LabVIEW (Laboratory Virtual Instrument Engineering Workbench) is a graphical programming language used as a powerful and flexible instrumentation and analysis software system in industry and academia. Cam r{ Camera stand Confined image acquisition chamber made ofblack polished iron sheet 39.4c 58.4T cm I 7. r _ ~ • Cylindrical lighting --~ 45.7 cm - arrangemerit Fig. 3.1. Schematic diagram of computer vision system 92 3.3.2. Image aegwsition{ vision system features The image acquisition is the process of obtaining a digitized image from a real world source and its step is generally comprises of illumination chamber and camera system (Fig. 3.1). Fruit is illuminated in illumination chamber and subsequently its six image in different postures (four by rotating 9D°, one from top and one from bottom, Fig. 3.2) were captured using four different camera set up and stored into the computer for further processing. Since, each step in the acquisition process may introduce random changes into the values of pixels in the image which is referred to as noise. The detailed description of these is given below. I.l Fig 3.2. Images of mango fruit acquired at different postures and respective segmented and equalized image 3.3.2.1. Illumination Chamber The black painted chamber fabricated from iron sheet has been used in this research work to protect the objects from the natural (or external) light and in order to provide diffuse illumination over the fruit surfaces, with the aim of avoiding highlights and light reflections. M, the image processing technique is very sensitive to illumination, the constant environment condition is very important to achieve a robust performance of the algorithm. The inside chamber of the illumination is controlled with the nine fluorescent bulbs (50watt, 220 Volts) fitted in cylindrical lighting system (Fig. 3.1) with electronic 93 ballasts and UPS system for constant supply of voltage (Fig. 33), This constants condition achieved by cylindrical lighting system was used for the visible light illumination of mango fruit in case of monochrome, color and NIR camera setup. En addition to visible lighting arrangement, the illumination chamber is also equipped with four 15W fluorescent UV tube (100nm-00nm) for the purpose of UY light illumination of fruits and capturing of reflected UV light image of mango fruit using UV camera. Although, before capturing of the fruit image using the monochrome, color and NIR camera setup, it assured that UV light was off and rice-versa. All cameras were fasted at 60 cm from the surface of object except NIR camera which was fasted at about 59 cm. Fig. 3.3. A glimpse of computer visionfimaging system used in this research 3.3.2.2. Cameras and Frame Grahber Four carncras such as Color (ROB), Monochrome, Ultraviolet and Near infrared were used. All cameras were mounted manually on the top of illumination chamber one by one at about 200 mm from the top of the fruit and images were acquired. The color }GB) and monochrome cameras were attached to PC via U B ports whereas the NIR and UV camera ware connected through Gigabit Ethernet port on the PC via Cat5e Ethernet cable and FALCON/ EAGLE Frame Grabber (Falcon PC1 board), respectively. The technical details of the different cameras that were used in image acquisition are given in Table 3.1. Fig 3,4 displayed a glimpse of UV, monochrome and MR cameras. 94 Table 3.1. Details of different cameras used for image acquisition of mango fruits S. Type of Model Serial N. Resolution Manufaiturad by/ No. camera No. made in I Visible R B U1- 4002718812 752 x 480 Il S imaging Color (u Eye) 1 2SLE- pixels development C-HQ system. Gemany 2 Visible Ui- 400275424 752 x 480 IDS imaging Monochrome 1220SE- pixels development M-GL, system, Germany R.2 3 Ultraviolet - 251447 768 x 494 Sony, Japan EUSO/06 pixels 4 Neal infrared N R- 00214 320 x 256 Vosslcohier Gmb}3, 300GE pixels Germany [A) (B) (C) Fig. 3.4. A glimpse of different cameras (A) UV Camera; (B) Visible monochrome camera; and (C) NJR Camera 3.3.3. Processing of images for physical Characterization At first, some image preprocessing operations employed on the input image (browsed and selected image) before applying the proposed physics' characterization process. Preprocessing steps are necessary for converting the captured images found from the real source so that they can be eligible for performing any binary operations onto it In these proposed methods, I have applied some preprocessing steps such as image 95 enhancement, noise reduction, etc. Again., at every preprocessing step there were a lot of options for using various kinds ofinethods in different appiications, 3.3.3.1. Monochrome Image processing The processing of acquired monochrome images start with the transfer of the image to the specific buffers of the RAM memory of the PC, where the image analysis process was applied. To cirry out the research wanks, different algorithms for Chausa and Dashehari cultivar have been written. For fast feature extraction and processing, images were cropped as per ROI (region of interest) using image viewer software (evaluation version). --11 1 .. . RR itiI_ - •— _'' 1 _ rsla hr A Fig,3.5 Shows histogram of original image (1), image after improvement of brightness (2), and of image after HIS-intensity plane extract kn (3) for Chausa and Dashehari fruits in case of monochrome camera. 96 In order to reduce the computational time and expense of the whole process the sizes of the acquired monochrome images of Cliausa and Dashchari cultivar are sub- mpled or trimmed to 435x29I and 4OS <252 pixels (m n) by nearest neighbour interpolation method, respectively. To enhance the visible monochrome image appearance of Chausa and Dashehari cultivar, the value of brightness, contrast and gamma was adjusted as 66, 44.50° and 0.27 (Fig 3.6b) and 66, 45.5O and 0.51 (Fig 3.7b), respectively, and to find out which color plane has highest contrast between object and backgrounl, histograms were developed before aM after brightiess enhancement of image. The developed histogram, in Fig. 3.5, after brightness enhancement showed that the intensity color plane of HIS can be used to threshold the object (both Chausa and Dos/zehurr) from the background. In oilier words, intensity color plane of HIS was extracted before the image segmentation. Then, the image segmentation was normally achieved by setting a threshold based on image histogram and the threshold was used to convert the underlying image into a binary image. The maximum threshold range for monochrome images of Chausa and Das ehari cultivars was adjusted manually as 65 (Fig. 3.ód) and 65 (r'ig.3.1d) for looking of dark object (back ground). The lookup transformations was used to highlight image details in an area containing significant information at the expense of the other areas after segmentation of image. The smoothing median filter of 15x15 size was used for removing the isolated pixels and blurred nature of equalized monochrome image. Thus obtained image was free from noises and hence used for the particle analysis to extract morphological features of mango fruits such as area, perimeter, Max Feret diameter, hydraulic radius, Waddel Disc diameter, elongation factor, compactness factor, Heywood circulatory factor and type factor. Further, the differentiation filter, a high pass filter, was used for the purpose of edge detection or boundary highlighting of the object in filtered monochrome (Fig. 3.6f and Fig. 3.7t) images of Chausi and Dosheharl cultivars. Edges are determined based on their contact and slope and are Located along multiple parallel search lines drawn within a rectang tar RO1 of the mango fruits using the clamp (rake-based) and caliper were finally used for measuring the distance between the first and last edges found. The horizontal max caliper was used to measures a distance in the ho ri ntaI direction from 97 th-; vertical sides of the 1 OI while to measure a disiunee in the vertical direction from the horizontal sides of the ROI toward the center of the•R I the vertical max caliper was used. The algorithms developed for Chas and Dathehari fruits are displayed is the Fig. 3.6 and Fig. 17, respectively. In this study, the. fruit length was measured by taking average of measured horizontal distances from the four profile images of a Mange acquired at 0°, 90°, 180° and 2700 postures (Fib. 3.2). Similarly, the width((at 0°, l00~, top and bottom) and thickness (900, ISO° and top and bottom) of fruits were also evaluated by taping the average (Fig. 3i). The geometrical attributes such as length, width and thickness of fruits measured by caliper arc also presented in the Fig. 3.8. 3.3.3.2. Color image processing After capturing The color profile image of fruit, images were browsed, selected and transferred to the specific buffers of the RAM menry of the PC, where the images were processed and anal' d using various IMAQ vision features, functions and operations. As discussed in monochrome image processing, color images were also cropped as per ROT (region of interest) using image viewer software (evaluation version) to reduce the, computational time. Thus, the sizes of the acquired color images of Chausa and Dashehari c.ultivar were sub-scrupled or trimmed to 466x369 and 4i9x277 pixels (mxn) by nearest neighbor interpolation method, respectively. The brightness enhancement, the next step, was applied to the images. image enhancement, actually, is the technique to make the image clearer so that various operations can be performed easily on the image. For this, histogram modeling techniques (often called histogram equalization) is a sophisticated image enhancement technique that attempts to modify the dynamic range and contrast in an image by altering that image such that its intensity histogram has a desired shape. Histogram equalization employs a monotonic, non-linear mapping which re-assigns the intensity values of pixeis in the input image such that the output image contains a uniform distribution of intensities (.e. a fiat histogram). This technique is used in image comparison processes (because it is effective in detail erahancenient) and in the oorrecticru of non-linear effects introduced by, say, a digitizer or display system. ` 98 [Original image: Browsed and Selected[ Bnghtnesc Enhancement all Color Planes [Brightness =66. contrast = 45.5-0 &y = 0.29] Plane Extraction Setup Threshold Setup: Manual Threshold f for dark object. min threshold range-65i . Luck up table: Equalization Filter I : Smoothing-median Tilter size (15x15)1 Filter 2: Edge detection —Diff. ParticIe analysis: [Kernel size (3x311 I I Measurements Clamp 1: Horizontal max. Caliper Area, Perimeter, Max Feret Diameter, Hydraulic Radius, Clamp 2: Vertical max. Caliper Waddel Disc Diameter, Elongation Factor, Compactness Factor, lieywood caliper: Distance measurement Circulatory Factor and Type between two points Factor Dimension (Length, Breadth Thickness) Fig. 3.6. Flow diagram of algorithm for dime sion and other particle measurements of Chausa fruits using monochrome CVS (a) Original monochrome image, (b) image after brightness enhancement, (e) image after HIS-Intensity plane extraction, (d) imams after threshold, (e) filtered image, (I) the outline of image, and (g) image with length and breadth measurements. 99 `Original image: Browsed and Selected] ` Brightness 1Cnh ncement all Color Planes I Brinhtness =66, contrast = 45.50 & v= 0.4111 Color Plane Extraction Setup F}1 I-Intcnsity Planel Threshold Setup: Manual Threshold 1 for dark object. min threshold range 51 1 Luck up table. Equa&atian I Filter 1: mouthing- median (Filter size = (15x15)1 Filter 2: Edge detection — Dif Particle analysis: MT .emel size (3x3)1 I easurements Clamp 1: Horizontal max. Caliper (Area, Perimeter, Max Feret I Diameter, Hydraulic Radius, Waddel Disc Diameter, Clamp 2: Vertical max. Caliper Elongation Factor, Compactness Factor, Heywood aliper: Distance measurement Circulatory Factor and `lope between two points Factor Dimension (Length, Breadth & Thickness) Fig. 3.7. Flow diagram of algorithm for dimension and other particle measurements of as!iehari fmits using monochrome CVS (a) Original monochrome image, (b) image after brightness enhancement (c) image after HIS-Intensity plane extraction, (d) image of er threshold, (e) filtered image, (t] the outline of image, and (.g) image with length and breadth measurements. 100 ianieter (i.e. Length of fruit) Intermediate diameter (i.e. Width of fruit) Minor diameter . (i.e. Thickness of fruit) meter of fruit Fig 3.8. Some geometrical attributes of mango fruit The Fig. 3.9 and Fig. 3.10, showed the histogram of all color channels (red, grcon and blue) before and after brightness enhancement of image of Chausa and Dashe/sari fruits, respectively. The brightness, contrast and gamma value of color image of both cultivars (hiusc and Lkisiehuri were adjusted as 44, 53.30° and 0.27 Fig 3.1Ib and Fig. 3.12b). After brightness enhancement, the blue, color plane was extracted because having bimodal of binary image. However, it was very hard to obtain proper bimodal histograms of images of all fruits because of lighting problem. From these figures, it is clear that blue channel of both cultivars has more proper bimodal than other color channels i.e. red and green. Image thresholding operation was applied after the blue plane extraction for the binari ation of image and the binarization threshold of image was estimated from the image intensity histogram. In other words, image segmentation was normally achieved by setting a threshold based on image histogram and the threshold was used to convert the underlying image into a binary image. In present study, the max threshold range for dark object or minimum threshold range for fore object was 99 for color image of 'hau!a cultivar (Fig. 3. lid) and 156 for color image of Dashehari (Fig.3.l d) enables to select ranges of ayscale pixel values. For the better results, the grayscale values of thus segmented binary image was equalized using LUT and then smoothing median filter of 1 5x 15 size was used for improving the binary image. Since, binary image generally needed some processing to 101 LL.4LL4 k. —R•~ Cl .nr.I — 4Fern Chen+dl1 —. BYa Cn.nn4 'cogs 4 7446 Al 10466 = 7 cae LL& - 8 so IaD 1810 144 220 3DC • d•] n!...Jy #mod Chneshal'— -~.en Q~ J —6Wo Oran 15606 B1 7COD0 II ~pp0 0 SOD 206 3m RAOI Frlarmny 0 0 P} 12C 750 2m m P- Frame 1{ig. 3.9. Shows histogram of original image (Al A2) histogram of image after improvement of brightness (B1& , and histogram of image after blue-RGB color plane extraction (C1& C2) for /iauga fruits in case of color (ROB) camera reduce the existing noises after LUT operation. The noise, in this stage,'is defined as the area detected as features other than fruits. Once the existing noises removed, the high pass dilTerentiaiion filter was used for detecting the edge of object in color image of fruits of both cultivars. This filter produces continuous contours by highlighting each pixel where are intensity variation occurs between itself and its three upper-left neighbors. The new value of a pixel becomes the absolute value of its maximum deviation from its upper-left neighbors. The kernel size that determines the horizontal and vertical matrix size was chosen 3x3. 102 LJ) J A A d inal f fl 1ar41 I 8Ja l7`an f~C~O A ~ 3.X70 1446 4 4 tw 200 3j Ftxal Yltmlr.Cv —FW O1omol —ec n Manr& — Bba Qunn ?IXIO Ii $ moo 4111 moo 20M low 0 50 100 154 200 290 300 8O 7D D a 5CJ1 4M 3]Ca 200 a 0 Fig.3.1Q.Shows histogram of original image {A,& A2), histograhistogram of image after improvement of brightness (R1 B2), and histogram of image after blue-RGB color plane extraction (C1& C1) for Dashehari fruits in case of color (ROB) camera Finally, the clamp (rake-based) and caliper function, works only with 8-bit rnachinc, was used to finds edges along a rectangular region of interest (ROI) by drawing in the image and measures the distance between the first and last edges found. In addition, various rnorphcligical rneasuremeits were also extracted from the fruits image, after just noise reduction, using particle analysis technique in IMA vision software, auto t tall}r. On the basis of above image processing steps, tht flow diagram of algorithms were developed for physical characterization of Chausa and Dashehari fruits, are given in the Fig. 3.11 and Fig. 3.12, resp actively. 103 Original image: Browsed and Selectedi Brightness Enhancement FBrightnes5 44. contrast = 53.30 &.y = 0.271 Color Plane Extraction Setup FRB- Red Piancl Threshold Setup: Manual Threshold Look for dark object, max threshold ramge=991 Luck up table: Equalization Filter 1: Smoothing-median [Filter size = (15x15)] Filter 2: Edge detection — cliff. Particle analysis: [kernel size (3x3fl Measurements Iamp 1: Horizontal max. Caliper (Area, Perimeter, Max Feret iameter, Hydraulic Radius, Clamp 2: Vertical max. Caliper addel Disc Diameter, longation Factor, ompactness Factor, Heywood Caliper: Distance measurement Circulatory Factor between two points and Tvree Factor I Dimension (Length, Breadth & Thickness) Fig.3.11. Flow diagram of' algorithm for dimension and other particle measurements of Chausa fruits using color (ROB) CV S. (a) Original RGB image, (b) image after brightness enhancerneiit, (c) image after blue plane extraction, (d) image after threshold, (e) filtered image, (f the outliiae of image, and (g) image with length and breadth measurements 104 Original image: Browsed and Selected Brightness Enhancement [Brightness 44. contrast = 53.30 & v = 0.271 Color Plane Extraction Setup ThresholdSetup: Manual Threshold [Look for dark object, max threshold range=156] Luck up table: Equalization Filter 1: Smoothing-median [Filter size = (l 15)] Filter I; Edge detection — duff. Particle analysis: [kernel size (3x3'f Measurements 'Clamp 1: Horizontal max.. Caliper (Area, Perimeter, Max Feret Diameter, Hydraulic Radius, Clamp 2: Vertical max. Caliper Waddel Disc Diameter, Elongation Factor, Compactness Factor, He} wood Caliper: Distance measurement Circulatory Factor and Type between two uaints Factor Dimension (Length, Breadth & Thickness) Fig. 3.12. Flaw diagram of algorithm for dimension and other particle riseasurements of Dashehari fruits using color (RGB) CVS. (a) Original ROB image, (b) image after brightness enhancement, (c) image after blue plane extraction, .(d) image after threshold, (e) filtered image, (f) the outline of image, and (g) image with length and breadth measurements 105 3.33.3. Ultraviolet image processing The processing of UV images were initiated by browsing, selecting and size trimming to 384x285 and 278x201 pixel of the Chausa and Dashehari mango fruit's profile image, respectively. The profile images were captured using LW camera when wits were illuminated with the IXV light (100- 400 nm). The brightness of such acquired images of both cultivars were improved by adjusting brightness, contrast and gamma value of only blue channel as 62, 55.0 and 0.46. The main purpose of brightness enhancement of image is to make the image more clear so that various operations can be performed easily on the image. For this, histogram modeling techniques (often called histogram equalization) is a sophisticated image enhancement technique that attempts to modify the dynamic range and contrast in an image by altering that image such that its intensity histogram has a desired shape. A B Fig.3.I3. Shows histogram of (1) original image, (2) image after improvement of brightness, and (3) image after intensity-HSI plane extraction of Chau. a (A) and Dashehari (B) fruits in case of UV camera Pfeil After brightness enhancement, the intensity plane with highest contrast was extracted for easy pros in of the image. As discussed above, before selecting and extracting the intensity ply of both fruit's image, hue and saturation planes were also tested one by one. Decision was finalized based on the clarity of usefulness and true information hold in the image. The histogr ms (Fig. 3.13) developed before and after brightness enhancement of image showed that the intensity plane of IBS can be used to threshold the object from the background. - In this current research, the maximum threshold range for Liv image of Chr usQ and Dashe ari cultivar was adjusted manually as 216 (Fig, 3.144) and 218 (Fig. 11 Sd), respectively for looking of dark object (back ground). Since, image thresholding is a segmentation technique which classifies pixels into two categories [(1) those at which some property measured from the image falls below a threshold and (2) those at which that property equals or exceeds the threshold]. There are two possible output values, thresholding creates and such created binary image was equalized by LUT operation. The smoothing median of 15x15 size was used for removing the noises such as isolated pixels and blurred nature of UV image, The particle analysis was carried out for extracting the several morphological features from the filtered image. The flow diagram of algorithms developed for physical characterization of Chausa and Dashehari fruits using UV CVS are given in the Pig. 3.14 and Fig. 5.15, respectively. Further, the edge detection of the object was performed by SobelSobe edge detector for extracting the outer diameters of the fruits using clamp and caliper equipped in the IMAQ vision tool after particle analysis of the filtered image. The Sobel operator does the first' order differentiation for detecting the discontinuities in the pixel intensity value. The 3 3 matrix Sobel operator was used for the edge detection in horizontal as-well as vertical direction. In other word, each pixel i5 assigned the maximum value of its horizontal and vertical gradient obtained with the following Sobel convolution kernels which highlights significant variations of the light intensity along the vertical and horizontal axes: —101 —1-2-1 Sx=-202 SY- 0 0 0 —103 1 21 107 1Original image: Browsed and Selectedi Brightness Enhancement of Blue ColorPlane [Briahtncss . contrast = 5550 & -+ = 0.461 Color Plane Extraction Setup rT QT_Tn+ari ;t r'D1nnal Threshold Setup: Manual Threshold iLook for dark object, max threshold ranEc= 161 I Luck up table: Equalizationn I Filter 1: Smoothing-median [Filter size = (5x15)7 Filter : Edge detection — Sobcl Particle analysis: Measurements Clamp 1: Horizontal max. Caliper (Area, Perimeter, Max Feret e Diameter, Hydraulic Radius, I Clamp 2: Vertical max. Caliper Waddel Disc Diameter, Elongation Factor, Compactness Factor, Heywood Distance measurement Circulatory Factor and Type Factor Dimension (Length, Breadth & Thickness Fig.3.14. Flow diagram of algorithm for dimension and other particle measurements of Chimsa fruits using UV VS. (a) Original UV image, (b) image after brightness enbancerneat, (c) image after H I-intensity Plane extraction, (d) image after thre hold, (e) filtered image, (f) the outline of image, and () image with length and breadth mcasuremonts 1,08 ]Original image: Browsed and Selected] Brightness Enhancement of Blue Color Plane I Mrinhrtncss =62. contrast = 55.50 & -Y = 0.46'i Color PIane Extraction Setup f lS I-Itttens its► Plan.el Threshold Setup: Manual Threshold [Look for dark object, nix ttueshold rane=2l81 Luck up table: Equalization Filter 1: Smoothing-median (Filter size = (15xl5)1 Filter 2: Edge detection — Sobel Particle analysis: Memel size [3x311 Measurements Clamp 1: Horizontal max. Caliper (Area, Perimeter} Max Ferot Diameter, Hydraulic Radius, Waddel Disc Diameter, Clamp 2: Vertical max. Caliper Elongation Factor, Compactness Factor, Heywood 'Caliper: Distance measurement Circulatory Factor and Type ` between two voints ` Factor Dimension (Length, Breadth & Thickness) Fig.3.15. Flow diagram of algorithm developed for dimension and rather particle measurements of Dashehari fruits using UV CATS. (a) Original UV-image, (b) image after brightness enhancement, (c) image after HIS-intensity plane extraction, (d) image after threshold, (e) filtered image, (f) the outline of image, and (g) image with length and breadth measurements 109 However, the Sobel filter assigns a higher weight to the horizontal and vertical neighbors' of the central pixels; `. 2 — P~j+1,1-L7 + 1 j_I p;+i,j1 `}` P(+-I,1+i) — max 1P _ — P + P 2 + P P 3.3.3.4. Processing ofNIR image Like monochrome, color and UV image processing, near infrared images were also processed and separate algorithms for Chausa and Dashehari fruits were developed. Flow diagrams of developed algorithms are displayed in Fig. 3.17 and Fig. 3.18, respectively, Browsed and selected NIR images of fruits of Chau a and Dashehari cultivars were sub s rmpled to the 99'c1 and $1x55 pixel (m n) size, respectively, in order to reduce compu[ation l time. Then, values of brightness, contrast and gamma of- A B IM Er kl I•: -& Fig.3.16. Shows histogram of (1) original image, (2) image after improvement of brightness, and (3) image after intensity-H I plane extraction of Chauxa (A) and Dashehw'i (B) fruits in case of NIR CVS 110 Oriel iiti .Bi'óédnd Slected ~r; IL d .r 1014 Brightness Enhancement of .1I Color Plane. Brightrtess =6 , contrast = 45.50 & j= 0.411 A.. Color Plane Extraction Setup [HS1-Intensity Plane] Threshold Setup: Manual Threshold [Look for hrght object, min threshold range=96] Luck up table: Et{ualization I Filter 1: Smoothing-median f Filter size = (1 5 15)1 Filter 2: Edge detection — Sobel Particle analysis- [Kernel size (x31 Measurements Clamp 1: Horizontal max. Caliper (Area, Perimeter, Max Feret Diameter, Hydraulic Radius, Wadde Disc Diameter, I Clamp . Vertical max. Caliper Elongation Factor, Compactness Factor, Heywood ip et: Distance me surem ent Circulatory, Fact-or and Type Factor Dimension (Length, Breadth & Thickness) Fio.17. Flow diagram of algorithm for dimension and for other particle measurements of C/R Wsa fruits using N [R CIS (a) Original NIR image, (b) image after brightness enhanceraerit, (c) image after H I-Intensity Plane extraction, (d) image after threshold, (e) filtered image, (f) the outline of image, and (g) image with length and breadth measurements 111 Fig.3.18. Flow diagram of algorithm developed for dimension and other particle measurements of Dasheharf fruits using NIR CYS. (a) Original NIR image, (b) image after brightness enhancement, (c) image after HSI-Intensily Plane extraction, (d) image after threshold, (e) filtered image, (#) the out line of image, and (g) image with length and breadth measuremexits 112 trimmed images were adjusted as 62.55, 55.5(' 0.41 to make the image clearer. For this, histogram equalization is applied that attempted to modify the dynamic range and contrast in such a manner that its intensity histogram has a desired shape. Histograms were developed before and after brightness enhancement to find out which color plane has highest contrast between object and background, Developed histogram of original image, image after brightness improvement and image after intensity plane extraction of both cultivars (Fig.3.16) showed that the intensity plane of HIS can be used to threshold the object from the background. Before selecting and extracting the intensity plane of image, hue and saturation planes were also tested one by one and histograms_ were developed Decision was finalized based on clarity of usefulness and true information hold in the image. The proper bimodal histogram, however, was hard to obtained for a]I foots images. The dark object (back ground) was looked and the bright object (fore ground i.e. fruit) was separated from the background by manual threshold operation by adjusting maximum threshold range as 96 for Cliauscr cultivar and 93 for Aashehari cultivar. Lookup table operation was employed to equalize the grayscale values for highlighting the image details in an area containing significant information. Although, the equalized image was found sometimes useful for direct detection of defect. The problems induced by light intensity gradients, poor contrast, irregular interfaces and touching blemishes needs to be solve, for accuracy of algorithm. The equalized image, thus, filtered using smoothing median filter of 15x15 size and Sobel filter (of kernel size 3<3), determines the horizontal and vertical matrix, is chosen for the boundary tracing of the abject. Finally, the clamp and caliper operation was used for finding of edges along a rectangular region of interest and for measuring of the distances between first and last edges. Similarly, the morphoIogical measurements were also done, before the edge detection, by the pathcle analysis technique which is already equipped in ]MAQ vision software. The morphological features extracted by the particle analysis technique using all camera set up have been discussed below in particle analysis section, 113 3.3.4. Particle analysis Particle analysis technique available in [IVfAQ vision software perforated on the processed images awl extracted nxiorphological features of the selected particIe, The attributes discussed such as area, perimeter, Max Feret diaraeter, hydraulic radius, Waddel disc diameter, elongation fietor, compactness factor, Heywood circulatory factor and type factor in this section attempted to describe the morphology of the outline and shape of the biological object. 33.4.1, Area The area of a binary object is found by counting the number of pixels in the binary object. This is calculated as the sum of pixels that are enclosed within the boundaries of the sample (MC. I997). It is then converted in calibrated units such as mm to enable direct comparison among sample images. The area of an object is a relatively noise-immtrne measure of object size, because, every pixel in the object contributes towards the measurement. In this study, the total area was calculated by the addition of area obtained from the profile image at 0° and 1800 postures of fruits. However, for the mass and volume modeling, the area obtained at 00 postures of fruits was only considers. 3.3+4.2. Perimeter The perimeter of a boundary defined by a list of ca-ordinates is the sum of the distances from each co-ordinate to the next. In other words, this is calculated the sump of the pixels that form the boundary of the sernple. It is also converted in caIibrated unit (mm). Perimeter offers a measure of size as well as a measure of outline roughness (Fig 3.8). The perimeter measurement can become distorted by the fractal nature of some boundaries (fine boundary detail may exaggerate the size of the measurerflent). However, assuming relatively smooth boundaries, the perimeter is a possibly desirable feature on account of its simplicity and because -of'the speed with which it may be extracted from a profile boundary (EC., 1997). In this study, perimeter feature of fruits Were evaluated and modeled for the quick estimation of fruits weight and volume. 114 33.43. Max Feret Diameter (ME)) This is defined the greatest distance possible between any two points along the boundary of the sample or the greatest distance between two vertical lines tangential to the ends of the particle. The MFT) obtained by particle analysis of. all images (monochrome, color, UV and NIR) were modeled for estimation of weight and vohume of fruits of both cultivars (NIC, 1997). 3.3.4.4. Waddel Disc Diameter WDD) Waddel Disc Diameter is the diameter of the disk with the same area as the particle (NIC, 1991 ). It is defined as: panicle area LID- (1) The D, thus, obtained from all four types of images was correlated to the weight and volume of fruits of both cultivars. 3.3.4.5. Elongation Factor The elongation factor is the ratio of the longest segment within an object to the mean length of the perpendicular segments (NIC, 1997). It is defined as: Max inter pI longation ,factor = - (3.9) Mean perp endi cu /ar int ercept The more elongated the shape of an object, the higher its elongation factor 33.4.6. Compactness Factor The compactness factor is the ratio of an object area to the area of the smallest rectangle containing the object (N1C. 1997 ). It is defined as: Campaciness . factor (3.10) Breadth x width The compactness factor belongs to the interval [0, 1]. The closer the shape of an object is to a rectangle, the closer to I the c mpactness factor. 3.3.4.7. Heywood ClrcuItory Factor The Hey ood cfrcu/arity actor is the ratio of an object perimeter to the perimeter of the ci'Ie with the, same area (NIC, 1997 ). It is defined as1 115 Par tk kperbneter- _ _ Far i dde perimeter Prirneterof circle withsarnaareaas particle` par1 ciearea 3.T I) The closer the shape of an object is to a disk, the closer the Heywood circularity factor to 1. r perfect circle would have a Heywood circularity factor oft, while a square weuld have a circularity -factor of 1.128. Langer particles would tend to have a high Heywood circularity factor, Thus, the Heywood circularity factor can be used to classify particles. 34.S. Type Factor It is a complex factor relating the surface area to the moment of inertia (NEC, 1997 )_ It is defined as: (particle area)? 'p factor = - — (3.12) 4J Where, I,.. and I. are moment of inertia with respect to gravity. 3.4. Image processing for defect detection Image processing artd analysis were performed by thresholding, filtering and ether preprocessing techniques {included it Lah iRW software}_ The 120 fruits (60 of each cultivar i.e. Ciausa and Das zart), including fresh and with various defects, were used for development of bruise detection algorithm using monochrome (Fig. 3.19), color (Fig, 320 and NIR. (Fig. 3.21) based iLputer vision system (CVS) as discussed above, Six images at diftbrent posture (Fig 3.) of each mango fruits were taken using camera based vision system and thus obtained ilO (360 of each cultivar) images of fruits from each system were processed and developed defect detection -algorithm. Three algorithms were developed using monochrome (Fig. 3.19), color (Fig, 3.20), and NIR (Fig. 3.21) camera based technology for defect Jet tiara on the surface of fruits of both cultivar. A reference image was taken and processed using NUVEAQ vision software to develop these algorithm. The various steps involved in processing of image acquired by monochrome and color camera during the development of algorithm are surnmarized in the Table 3.2. Fig 3.21 showed the flow chart of an algorithm developed for the quantization or cal cation of percent area of black spot underneath of the peel in tern s of defected area I16 using N]R camera.. This flaw chart represents various image processing steps used in algorithm during processing of images using IMAQ vision system and the bruise detection and classification method is based on thresholding the gray levels of mango image pixels. Each image is preprocessed in terms of brightness improvement, contrast enhancement and setting up the specified value of gamma to get more information and easiness in thresholdi.ng. Table 3.2: Processing steps used in processing ofimages for Segmentation of defe is S. Techniques used in Image acquisition system No. image processing Color Monochrome 1. Original image Browsed and selected Browsed and selected Brightness Brightness: 78, Contrast: Brightness: 58, enhancement 58.60 and iamma (y): 0.63 Contrast: 53.30 and (}amrna (y): 0.59 3. Color plane Extraction HSL: Hue plane HIS: Intensity plane setup 4 Threshold setup Automatic threshold: Metric Local threshold: (Miry: 91) Background Correction with Kernel size (WxH) 87x63 5 Application of Lookup Equalization of image Equalization of image Table (Redistribution of pixel (Redistribution of pixel intensities) intensities) 6 Cray Morphology Dilation of image, Size: 3, NA Iteration: 1 7 Filter Smoothing Median Filter, W Size: 4>4 8 Advanced R oval of small object, Removal of large and Morphology Iteration: I small abject with Iteration: I 9 Binary Inversion Reverses the dynamic of an NA. image containing two different grayscale populations. 10 Particle Analysis Displays measurement results Displays measurement for selected particle results for selected measurements performed on particle measurements the image. performed on the image. 117 Original image Browsed and Selected Brightness Enhanced Brightness X58 Contrast = 53.30 Gana (y) = 0.59 Color Plane Extraction Setup HSI-Intensity P]ane Local Threshold: Background Correction Looked for dark object Kernel site (W'H), 87x63 Performs background correction to elniinate nori- uniform lighting effects and then performs threskiolding using the interclass variance thresholding algorithm. Luckup table: Equalization Adv Morphology (1) Removal of large objects Iteratiosi =15 Adv Morphology (2) Removal of small objects Iteration =1 Real , i 3 + 5 ku ,WLQ 'o I31a,+]t~C~4 B7,p0 U Fig. 3.19. Algorithm steps for defect detection on surface of Chaisci and Dashehw i cukivars using monochrome CV S 118 Original image Browsed and selected from storage drive in computer hard disk Contrast enhanced image after setting The brightness value: 7, contras: 59.60 and g- (Y): 0.63 Color Plane Extraction FISL: Hue Plane (Hue color plane of image after color plane extraction) Threshold Setup Automatic threshold: Metric, (Looked for bright object with minimum threshold value as 91) Image. equal ation using Lookup Table Gray Morphology I Image dilated with stitictured element siz, 3 and iteration as 1 Contd.... 119 Filtered image Smoothing Median of 4x4 size was applied for filtering the image Advanced Morphology Small objects removed Iteration: 1 Gray Morphology 2 Image dilated with structured element size 3 and iteration as 1 Binary inversion Particle analysis P. lb.., 1 2 Perime[er 1042.25531 1252O Holes Pedrne tr ii 297.35573 0.000000. Area II 73916.000 13 OC000 llale5' Area 1Q51OoDoo d.000LD0 Pastcle a ki, es Ar 47Oat2QJ 1QOUD Fig 3.20. Flow diagram of developed algorithm for defect detection on the surface of fruits of both cultivars (hausa and Dashehari) using color (O13) DVS 120 iII ______• Iq (1) Original image etup: equalizatio ri :;- I. ------r ----- t - i- (2) Contrast Dnhancement (6) Particle filter: Centre of mass II... tI- ri rrrTh, J. - (3) Color oparator (7) DefeiEecf area Subtraction F- 1•-• i I r (4) Color thresholding ) Particle Analysis: defected area usine RGB Model ______measurement Fig 3.21. Flow diagram of algorithm step s for defect detection underneath the fruits surface/Feel using NIR CVS 121 After image improvement, subtract color operator was used and image was subtracted with the value of hue, saturation, Mina, red, green and blue as 180, 240, 60, 64, 0 and I2, respectively and then tk resholded by some specified value of R.GB color model [R (0, 5), (0, 112) and B (0, 9)]. The lookup table was used to equalize/redistributes pixel intensities in order to provide a linear cumulated histogram and the Particle filter was used to extract defected area by using centre of mass parameter ranged between 2 and 3$_ After filtering the defected particles, the number of pixels were counted whose gray level is lowered than the threshold using particle analysis step of IMAQ vision software. In contrast to monochrome, visible (RGB) and NIR camera based imaging systems; various types of defects on fruit's suffices were also detected using the UV cunera based computer vision technology just after the image acquisition and brightness erxhancement. Five bands pass filters of 200 nrn, 2 0nm 3{lOnm, 350nm and 400nm wavelength were tested for defect or bruise detection an the surface of mango fruit by capturing their UV images. The result, thus, obtained have been discussed in the chapter results and discussion; section 4D. 3.5. Data analysis Measurements of all properties were triplicated. Maximum, minimum, coefficient of variation (CV), etc. were obtained using Microsoft Excel (2003) software. Multivariate analysis were also carried using SPSS software (version 16) to check the effect of main factors (harvesting stages, H; storage periods, SP acid cultivars, V) and their interactions (H* P, BV and SP y) for all two ways on physico-chemical properties of selected mango varieties. The paired t-test and the mean difference confidence interval approach were used to compare the length, width and thickness determined by the IP technique (applied for images acquired by different image acquisition system and lighting arrangement) and the actual (DVC) geometrical dimensions (L, W &T). The paired t-test was used for testing whether the difference between two meaurement5 were igriifieant or ztvt. The important feature of this test is its ability to compare the measurements within each subject. inally, the Bland-Altman approach (Bland and Altman, 1999) was used to plot 122 the agreement between the calculated and measured fruit dimensions. The statistical analysis, paired t-test and Bland-Altman plot, were all performed using Medcalc software (USA/Germany). The other software that used in the data processing and gearnelrical modeling was Unscramb'er 8,0.5 (C MO AS, Trondheim, Norway, version 8.0.5). 3.5.1. Geometrical attribute based modeling of fruit's weight and volume Several models were developed based on geometrical attributes (length, width, t i kness, aspect ratio, AMD, G-MD, sphericity and other multiplicative attributes obtained frorn length, width and thickness up to third order) of mango fruit obtained from image processing technique. Different cameras (monochrome, color, UV and Nom) and lighting setup were employed, as discussed above, to acquire the image of the fruits and then processed using Lab View software. Geometrical attributes, thus, obtained from these different tyre of images (monochrome, color, 1W and NI) were modeled for the prediction of weight and volume of flits using Unserarnbter softw re. Calibration is the fitting stage in the regression modeling process. The main data set, containing only the calibration sample set, is used to compute the model parameters (PCs, regression coefficients). For weight and volume various different calibration models were developed. The quality of the caIibiation models were quantified by the standard error of calibration (SEC) (eq. 8). the standard error of prediction (SEP) (eq. 9) and the correlation coefficient (R) between the predicted and the measured parameters. Thus, good models should have lower SEC and SEP. higher correlation c f dents but smaller differences between SEC and SEP1 because large differences indicate that too many latent variables are introduced in the model and the noises are also modeled (He et al. 20t)5). SEP ~ (~; — y1 —bias)1 (3.13) SEA'= F ( r YO )7 (3.14) I—I bias (3.15 ) p 1.1 123 Where is the predicted value of each observation and y is the measured value; 1 is the observation number in the calibration set and 1F is the observation number in the validation set; and bias is the systematic differences between the predicted and the measured. 3.5.2. Performance evaluation of algorithms The performance of algorithms developed for monochrome and color based computer vision system was evaluated with respect to manual assessment by expert. These sorting systems (monochrome and coIor based) were used under condition considered to be optimum with regard to the lighting arrangement and manual sample placement rate. Fruits (100 for monochrome and 180 for color camera based vision system) of various degrees of defects (fresh or slight defected, very severely defected, severely defected and minutely defected) were considered and sorted in four groups as per iefects perzznts with the help of expert The performance evaluation of proposed algaritI ms were carried out by calculating its accuracy, efficiency and processing time as discussed below, ii Accuracy ; 100 3.17 Where 'n' is the number of defected fits detected by image processing techniques accurately and the N' is the number of fruits detected manually E cie c}y= ~ 1aO 3.18 Where 'pn' is the number of fruits processed by techniques and 'pN' is the total number of fruits used Over all time taken by images was calculated by using performance evaluation meter in I1MAQ software automaticall y. 124 Chapter 4 Results and Discussion 4.1. Studies on variability of physicu-chemical parameters of mango fruit 4.1.1. Dimensions of fruit The mean values and standard error in estimation at every harvest stage and for each cultivar displayed in Table 4.1.1. The mean value of fruit length varied from 10064 to 101,87 mm between the three harvest stages and the Table 4.1 ,1 shows significant variation in fruit length among the cultivars at the 5 % level but not found significantly different among the three different harvest stages. Although, interaction between harvest stages and cultivars (HS*V) had significant effect on frutit's length at least at 5% level (Table 4.1.2). Table 4.1.1 shows that Chousa (112.32 mm) is longest fruit among the cultivars which is in agreement with the reports of Nandani and Oommen (2002). The n ximum length of ChausQ fruit wis found to be 131.2 mm in this study which is almost similar to the value reported by Nandani and Oommen (2O2) i.e. 134 mm, Furthermore, minimum and maximum values of average fruit length were also recorded at pre-mature stage of Hallo (89,44 mn) and Chausa (114.56 mm) cultivars (Table 4.1.2). The minimum mean value of fruit breadth was found to be 53.05 mm of the Dayhehari fruit harvested at pre-maturity stage while maximum mean value was recorded of the Chausa fruits harvested at the over maturity stage (Table 4.1.2), and the effect of interaction of harvest stages and cultivars (HS*' was significant on the fruits breadth. In addition, the effects of harvest stage and cultivars were found highly significant at the 5 % level (Table 4.1.1) and varied between 60.45 turn, and 64.00 mm mean values during pre- to over maturity stage. The width of mango was highest for Chausa (67.56 rnm) followed by Langra (66.09 mm) and MaFdQ (59,29 rum) and then Dashehari (55.53 mm) cultivar (Table 4.1.1). The Appendix-1 (A- 1.1 to A-1.17) show the tables of Analysis of Variant_ 125 fable, 4,1,1 b 'cd ofha st tae nd ciltiv r as same impodwphyiit1 POMM I& of mugu frail 1~ur~ Iromiters r th Rreedtk '1'hirkne4 1I) (ml) W pht ,) 'olurtic (m he Apl[tr~tio 1, Rab at stags 521b ? r thturc (Hi I 10,647' ).49" 5]r1 71.4311 !834 ]S472 O5 Ia ti 914 t U,5O6 ±449 f x 9} 1 4 y4.35$ Optimum matmc 1 O1,$C9 6I74 f 7 417 7.772 ` 2 357 211.56' (1111 = O.P4 0.46 0.44 :J D. 35@ } 0.518 14.368 f4.J Ov'tfm~tutt (Hi) 141.E l 1Y28l S .401 74.7 72.629' 2363?b1 11.059 0,550 10489 10'9 105 1477, 4f 2, Cu1Ii ' r 'h EU 1113V 56 kIIV U(1.4 ' 175U y 171.31t +I. }O.5I1 iDA5 (J, 3 14,777 l h arl 1].1 53~ , S .4n, io sr &r49 1V7i5J~ 110.4` 0a, + 31 ± U.484 14.4311 ±')527 .tJ45 f4,19~ f 4.1{15 10,014 L 'ru (V;) 9I I0% 5S9` 71. 11 3 3~1. 13 1b UJ1331 ± 1,2 9 111.674 . (599 t 0.734 14,69 1 x.86 U l # a1 911'' L),29h 54 S.R' I66.51 [9I 14 OI I J:. .29 '1. L,674 f 0,005 S r I IL' ..1 gM Winn I I3 IY 151.5311 5'x.21 '14.1$43 65.4196 2O.48 J [t44 StaudL9 d M (1) 0,7444 o681 1454 0,1954 0 514 3Ifl 2 3 7~ T lel f ni36 343 3 c)3 3I3 _ }3 mm si r l~f wit and i*r ~~ ni nt at I cast S~#~ l el U6 Table: d.I,2 tcrnibncd ffc of lirctDr IH[vcgt sLgc (I} and cgltivar ), on s me ern nt A*,j p 'imjirs fm~1po Combiot Lgnglh (mm) givat Thick ID Gl [1 obt YQ1u± pberkily Aso d f for (rim) (En) (r h1) {mm} _ (ml) ra1i IHlV1 1]4 1-1,59 66.47 O.8+ 57S.4,?3 ?~,5 II 7III5 23949 717 155 67,U1 0,6640,0X6 &,11480 H I V2 102.6011 59 515 ;1)J ] 47.19 O 73 67,6 I.89 63. US 139.451?.I 7 I45i Q17.O1 21 r11Oi 1,9 1I.R1 R1V3 9i9 11. 4. 11,1711+1.114 '1.±l21 ).82 L2 194.43#10,1.11%]10,)1 01721±4 U9 66i2±lr1O HIV4 11' 4 12.'Li 117 .341.14 i.5i27 &I, 2015.1f1U.i 1572±9,9i I1.734 019 (i lF1,10 H2VI [email protected] {1.59 64.$14 x,83 51 X0.73 7.7 } 0 $9 +4,6 U. 141. + 1I1714?, ~}7 01 'O. $ x.446 59 4}.80 .2V2 14.52 'I:1,59 57I .83 5 i273 ?1,81 .89 7.6]I , 5 174.4117. I? !77.22±7I0! O.241O. 5I.±O B0 H2V3 1 1,01±2. 65.%± 117 iOl±1114 7 .7L27 73,5 1,21241.07f1O1 371 ,91 3JJ 6{,±110 H2V4 4h1.2 60.5011.17 4((1.27 7Mii I.2f 1 17 4 ± D. I3 14.17.I 0.7% 1.009 f l~.511.10 1I3V1 112,9 1-1,3911,17 424 l3.i.127 I,9 1,2C' 312,4?1[0.13 32 LIi9 i 07 ' I ~9 113 Vi 1~717±h 6 5r16±@ 52.17 .77 7216±O,4 8.353.0 15 ?1=?.4S : 9,J±7.32 064114.016 51. Ii3V3 11 . -k2.25 6. 'I: 1.17 1r741X4 7 .99 1127 75 I f 12 14.74-'l0.13 261 67 9.91 0,7460009 , 11,10 U3V4 4.+2.25 X9.03-1.17 S5 1--I ,] +127 (61 ].1i 10 .$ -10.13 :3 991 o.1- o.O4 4.9=1,10 Si silica U U $1 II U III tM A Ieve] F t cg)! fie t aid ]emt Rt ] 141, El - pre, nplimum & ma llsn t ' V] V2, V& V4- CJiu.c4 Dr i rI, i&Lanii wIturvar hrmst s LV Similarly, fruit's thickness among the cultivars (CJ aura, .ache ari, Langra and Malda), and three harvest stages (pre-, optimum and over maturity) found to be varied significantly at the Level of 5% ( 4.O5) (Table 4.1.1). The interactions bet'een harvest stages and cultivars (HS*V) were highly significant (p<0.05) at the 5% level therewith in relation to thickness of fruits (Table 4.1.2). The Table 4.1.1 displayed the mean values of thickness and error in estimation at every harvest stage, for each cultivar and overall grand mean values (57.21mm), and grand mean error in estimation (1.O9mm). However, Table 4.1.2 presented the mean values of thickness of fruits of each cultivar harvested at pre-, optimum and over maturity stages. The Chausa fruits harvested at over maturity stage were found thicker than the other cultivar's fruits harvest at any stage of maturity..However, the reason of increase in fruit's width and thickness from pre.-maturity to over maturity might be due to more appearance of auxins that play a major role in the growth of the fruits. 4.1.2. AIM]) and GMD or mango size The effect of harvest. (pre-, optimum and over) stage (Table 4.1.1), cultivar (Table 4.1.1) and their interactions (Table 4.1.2)) on the size (GMD) and arithmetic mean diameter (AMD) of mots, were found to be significant (p<(L05). There is a significant (p 128 The major axis has been found to be indicating the natural rest position of the material and hence in the application of compressive force to induce mechani c rapture. AIso, this dimension will be useful in applying shearing force during slicing (Owolarafe and Shotonde 2O4). 4.1.3. Fruit weight and volume The variation in fruit's weight and volume among the cultivars and harvesting stage from pre- to over maturity -stage were highly significant (p The effect of interactions (Table 4.1.2) between harvest stage and cultivar (fi * on variability of fruits weight and volume was found highly significant (p 129 Table: 4.1.3. Sutrunary of effect of harvest stage, cultivar and storage period on TSS, titratable acidity, total carotenoids, specific gravity and colour values (L. a & b) of mango fruits Factors Parameters Total Soluble Titratable Total Specific Colour values Solids (°brie) Acidity, % Catotenolds gravity L a 1, . =1100a 1. Harvest stages Pre-mature (HI) 14.2528c 1.353a 0488b 0.978b 45.89c -7.329c 25.948c Optimum mature 15.62Thb 0.980b 0,47 6b 0.994b 43.17b - .73Db 28.395b Over mature (Hi) 16.6903a 0.70 a 0.597a 1.019a 50.60a -5.615a 33.624a Standard error (±) 1.354 0.046 0.021 0.008 0.507 0.199 0.445 Siitieant level ## * ** ** 2. ultivar Chausa 13.261 lb 0.613c 0.307c 0.9811 48.86b -7.539b 25.406b Da rehari [4.0056c 1.171a 0.681a 0.980b 45.30d -5.772a 31.351a La~tgra 15.3685b 1.299a 0541b 0.996b 51.98a -7.620b 29.855a 1l afda 19.4593a (.969b 0.554b 1,032a- 46.74e -5,301a 30.677a Standard error (±) 1.564 0.053 0.025 0.009 0585 0.233 0.514 Significant level ** ** ** Storage periods 0 days 9.O$ lbf 2011a 0,156 0.945c 40.702f -9,342d 28.435b 3 days 1 .342e 1.538b 0.230e 0.962e 46.065e -3.702d 28.283b 6 days 14,5444d 1.026c 0.372d 0.998b 47.902d -7.787c 28.804ab 9 days 17,2694c 0.673d 0.562c 1.012ab 49.666c -7,235c 29.146ab 12 days 19,2056b 0.479e 0.728b 1.023ab 51.061b -4.178b 30.518a 15 days 20.694a 0.352e 1.074a 1.043a 53.91 Sa -2.106a 30.696a Standard error (±) 1.915 0.065 0.030 0.011 0.0.404 0.282 0.630 Significant level ** Grand mean 16.619 1.013 0.521 0.997 48.9-19 -6.555 29.322 Grand mean 0382 0.026 0.012 0.004 0.165 0.115 0.257 Statidard error (f) Total no of fruits 216 216 216 216 216 216 2,16 Alphabet letter shows significant or insignificant relation **~ significant at least at 5% level 130 4.1.4. Sphericity Table 4.1.1 displays significance (F (21O5) 56.263, p = 0.000) variations in the sphericity of fruits among the four cultivars and also found to be increased significantly (p<0.05) from the pre- to over maturity stage of harvesting at the level of 5%. At pre-maturity stage, the average sphericity of tha mango ftuits (all cultivar) calculated as 0.685±0.004 which increased to 0.7 2*0.U04 at the over maturity stage of harvesting. Similarly, among the cultivars, ,4 ida cultivar fruits were evaluated highest in average sphericity value (0.754±0.005) followed by Langra cultivar fruits (0.733±0.005. In addition, the effect of interactions between harvest stages and cultivars on sphericity were also analysed and found to be significant at the confidence level of 5% (Table 4,1 2). In contrast, from the Table 4.1,2 it is also clear that the variation in the Dashehari fruit's spbcricity (i.e 62.10% at pre-, 62.4 % at lopti min and 64.10 % at over maturity stage) is not much Considerable. Furthermore, the grand mean value and mean value of sphericity of fruits of each citltivar is given in the Table 4.1.1 while the mean value of each cultivar at different harvest stage is dispIayed in the Table 4.1.2. These results, thus, seemed to be confirmed the information reported by Tha et al. (2006) for Dashehari cultivar in which they showed that an increase in sphericity was not so substantial? and only a slight change was noticed during the growth. The value of sphericity of Dashehari fruits reported by them i.e a 67% to 70% differs from the current findings might be due to the environ factors and orchard location. In addition, the sphericity of these cultivars was found higher than the reported value of orko (64.00%) by Owolarafe and Shotonde (2004) and m €rile (5$.32%) by Aydin and Ozcan (2007). And found lower than the reported value of gumbo (77.804) by Akar and Aydin (2005), strawberry (81%) by Lorestani and Tabatabaeefar (2006, wild medlar (0.90), sweet cherry (85.27%) and cactus pear (83,10%) by Haciseferrogullari et al. (2005), Vursavus et al. (2006) and Kabas et al. (2006). 131 Table: 4.1.4. Summary of effect of interaction b tween harvest stage ) and cultivar (V) ott TSS, titratable acidity, total carotenods1 specific gwvity and colour values (L, a & b) of mango fruits Combined factors Parameters Harvest stage Total Soluble Titratable Total Specific L a b H)* 1tivar (V) Solids Acidity aroteuolds gravity HIV1 11.861 0.697 0.455 0.932 53.288 -7.508 24214 E11V2 12.022 1.831 0.521 0.964 39.938 -6.434 27.711 H1V3 15.433 1.539 0,536 0.994 4.573 -8.445 23.155 H1V4 17.694 1.344 0,441 1,022 43,752 -6,928 28.70 V1 14.389 0.656 0.303 0,999 45,691 -7.703 22.510 H2V2 14.339 0.999 0.620 0.974 47.41-6 -5.752 31,592 L-12V3 14.9.2 1.401 0.495 0.981 52.794 -7.750 29.651 H2V4 18.81 0.865 0.486 1.031 47385 -5.717 29.824 H3V1 13.533 0.487 0,162 1.020 48-,210 -7,407 29.489 H3 V2 15.656 0.683 6.902 1.003 48.537 -5.130 34.746 H3 V3 15.750 4.656 - 0.91 1.012 56.564 -6,666 36.760 H3V4 1.8?. 0.699 4.735 1.042 49.079 -3.25 33.500 Standard error () 0.218 0.092 0.043 0.0 .5 0.512 0399 0.591 Significance level NS ** ** ** non. si nificatdt and **- sigriifsoat at lent at 5% level 2H1. H2 & H3 - pre, optirnum & over harvest stage 3'V1, '12, \3& V4- Cha#isa, Dash ari, Milddu & Lmr rn ltulva r hnry St stage 4.1.5. Aspect ratio To analyze the fruit whether roll or slide on their flat surfaces, the knowledge of aspect ratio is very important. If the value of the aspect ratio is being close to the value of sphericity, the fruit will undergo a combination of rolling and sliding action on their flat surfaces (Razavi and ?arvar, 2007). The grand mean vab es of aspect ratio (Table 4.11) of fruits computed as 60.39 % at 95% CL The highest ispect ratio (which relates the, fnit width to length) value of 132 Maldai cultivar harvested at over maturity stage (69.9) indicates that the fruit might follow the rolling than sliding action on their flat surfaces (Table 4, ii). But, if the value of the aspect ratio is being close to the value of sphericity, the fruit will undergo a combination of rolling and sliding action on their flat surfaces (Razavi and Parvar, 2007). Further, the value of aspect ratio found to be changed significantly (p<0.05) from pre- and optimum maturity stage to over maturity stage, However, there was no significant variation recorded in aspect ratio of fruits harvested between pre- (6(L6%) and optimum maturity stage (61.4%). Similarly, the effect of cultivars on aspect ratio of fruits was noticed highly significant (p<0.O5). Though, the interactions bctween harvest stages and cultivars (HS 'V) were found effected insignificantly at 5% confidence level (Table 4.1.2). 4.t6. Specific gravity The significant (p<0.05) effect of harvest (pr-, optimum and over matured) stages, cultivars and storage periods was observed on the specific gravity of mango faits (Table 4.1.3). The mean value of specific gravity varied between 0.97 glcm to LO 19 g/cm 3from pre- to over maturity stage of harvestin , respectively and among the cultivars 0.980 /cm3 to 1.032 glcin3. The average specific gravity of Maldai cultivar was evaluated highest i.e. 1.032 g/cm3 (Table 4.1.3). Similarly, the specific gravity varied significantly (p .05) between 0.945 cm3 (at 0 day) to 1.043 (at 15th day) during the storage of mango fruits. However, the effects of intesactforxs:.1 * (Table 4.1.4), SP* l (Table 4,1.5) and HS* SP (Table 4.1.6), observed insignificant (p>0.05). Table 4.14 displays the mean values of specific gravity of each cultivars with respect to every harvest stage. For instance, Maldai cultivar has highest specific gravity (L042 gkzn3)harvested at over maturity stage. Table 4,1.5 is showing the variation in specific gravity of cultivars with respect to storage periods while Table 4.1.6 shows variation in specific gravity of fruits with respect to harvest stage and storage periods. The maximurii specific gravity (1.089 u/cm) was investigated at 15`h day of storage of Maldai fruits white minimum value (0.901 g/cm) was recorded at fresh condition of Chausa cultivar (Table 4.1.5), On the other hand, the Table 4.1.6 showed that the fruit harvested at over maturity have highest specific gravity (1.057g1cm3) at 15th day of storage. 133 Table: 4.1.5. Summary of effect of interaction between storage period (SP) and cultivar (V) on T S, thratable acidity, total carotenoids, specific gravity and colour values (L, a & b) of mango fivi#s Combined factors Parameters Storage periods (P)C Total Tltralabl Total Specific L a b Cultivar (I) Soiublt e Acidity CaroternkIs gravity Solids_ S1 V1 7.061 1.010 0.009 0.901 40.131 -9.60826.931 S 1V2 8,644 2,411 0.157 0 933 39.665 -9,742 2 613 S 1V3 8,989 2.482 0.187 0.913 44.91I -9.923 28.918 S 1V4 11.644 2.140 0,193 0.981 38.101 -8.096 29.475 S 2V1 10.600 0.740 0.127 0.954 47.059 -8318 25.649 S 2V2 1-0.956 2.041 0.254 0,957 42.060 -9.521 29.786 S 2V3 10.911 2.055 0.263 0.980 50.067 -9.202 30.286 S 2V4 16.900 - 1.316 0.277 1.009 45.074 -7.568 27.410 S 3 1 12.167 0.633 0.197 1.000 48.343 -8.033 25.386 S 3V2 13.044 1.137 0.480 0.980 44.692 4.762 28.406 S 3V3 13.544 1.29 0.298 0.990 52.090 -8.240 28.740 S3 V4 19.422 0.903 0.514 1.022 46.484 -6,111 32.683 S4V1 14.900 0.512 0.271 1.003 50.663 -8.051 24.824 S 4V2 15.300 0.731 0.750 0.989 46.077 -7.333 32.549 S4 V3 17.244 0,825 0.511 1,016 53.I4 -7.878 28.523 54V4 21.633 0.626 0.657 1.041 48.781 -5.677 30.686 S 5V1 16.689 0.472 0.346 1.007 52.159 -6.936 25.251 S V2 17.078 0.392 1.039 1.002 47.179 -1.369 33.352 S 5V3 20.233 0.590 0.820 1.028 55.047 -6.281 31.812, S 5V4 22.822 0.462 0.707 1.054 49.859 -2.127 31.658 S 6V1 18.144 0.313 0,811 1.019. 54.823 -4.090 24.353 S6V2 19.011 0.314 1.404 1.022 52.111 2,095 35.394 S 3 21.289 0.412 1,104 L047 56.606 -4.200 30.854 S6 V4 24.333 0.369 0.976 1.086 52.133 -2.727 32.152 Standard error (±) 0.308 0,130 0.060 0.022 0.808 0364 1.260 Significance level '- "9 non signinicant and **- sigWficant at least at 5% level - 2s1, 52183, S4, S5 &. 56 - Storagc period at interval o13 days from the fresh condition ' 1, V2, V1& V4- Chausa, Dashehari, Makc i & Langra cultuivar harvest stage 134 The above findings are in close conformity with the result reported by Zaied et al. (2007) for ripe fruits of Chausa cultivar. Furthermore, the specific gravity could be tested as one, maturity index' (Ketsa et al. 1999) from many maturity indices. But, there is #o consensus on maturity indices due to differences among cultivars, production conditions and locations ( itra and Baldwin 1997). Harding et al. (1954) concluded that specific gravity cannot be used as a reSiable maturity index. - 4.1.7. Total soluble solids Table 4.1.4 displays significant (p Further, Ma/dui fruits were investigated highest in TSS content i.e. 19.46 brix among the cultivars. Table 4.1.4 shows that Makki eultivar omits have ISS content highest at over maturity stage of harvesting and at the 15'~ day of storage period (Table 4.1.5). However, the Table 4.1.6 reveals that fruits of all cultivars harvested at over maturity stage have maximum TSS content at 156 day of storage, The increase in TSS content in every cultivar might be due to the alteration in cell wall structure and break down of complex carbohydrates into simple sugars from pre to over maturity stage. The increase in sugar content was accompanied by a decrease in acid and increase in carotene. Results showed in this research are a. agreement to the Jha et al. (2006) that the fruits have TSS below S Brix are immature and TSS at 8 Brix or more than 8 Brix and about I ° o acidity could be taken as indices for full maturity for some varieties of mango. Popenaa et al. (1958) also had determined chainges in soluble solids content (from 8.0 to 11.0 % in hard fruit 135 and from 9.5 to 1&0 % in soft fruit) in °Haden' mangos and Harding of al. (19 4) found from 7 to 10 lo in mature-grcea `Haden', ` eitt', and `Kent mangos and this rho increased with maturity and ripeness (14 to 16%). On the other hand, Baer-Sanudo et aI. (1999) measured soluble solids contents of `Haden', `Kitt', `Kent', end `Tomrny Atkins' mango cultivars 7.3, 6.6, 7.4, and 7.3 %o at the stage when its flesh was cream (white) i.e. completely immature. 4.1.8. Titratable acidity The effects of harvest stage, cultivars and storage periods on titratable acidity of the mango fruits are presented in the Table 4.1.3 while the combined effects of harvest stage and cultivar, storage period and cultivar, and harvest stage and storage period on titratable acidity are presented in the Table 4.1.4, Table 415 and Table 4.1.6. These tables reveal that all factors (harvest stage, cultivars and storage periods) and their, combined interactions have significant (p<0.05) effect on titratable acidity. For instant, the mean value of TA of fruits decreased significant (p<0.05) from pre-maturity (1.353%) to over-maturity (0.706%) stage of harvesting which agrees with the previous values reported by Araiza et al. (2005) that titratable acidity declined (from 120, 0.81, and 1.$9 % in mature-green `Ataulfo', °Kent, and `Tommy Atkins', respectively,) about 0.5 or lower when ripen. However, in developing fruits acidity increased at early growth phase but reached a peak and then declined gradually until harvest (Wardlaw and Leonard 1936) and again decreases with ripening (Gowda and i(amanjaneya 2001; Ueda et al. 2(101; Tandem and Kalra• 1986). This finding is also an agreement to TJpadhyay and Thpatha (1985) and Medlieott et al. (1986) that the TA decline during the later stage of growth on attainment of maturity and ripening. _ Similarly, as the storage period increased, the mean value of TA of fruits decreased signiticantiy and recorded lowest (0.352 %) at 15th day of storage. In addition, a significant (p<0.O5) variation in TA among the cultivars was also recorded and Lc gra cultivar fruits were evaluated sourer than fruits of other cultivars followed by Dashehari cultivars (Table 4.1.3). Further, from the Table 4.1.4, it is clear that the Chausa cultivar fruits have lowest titratable acidity (0.487%) or sweetest at the over-maturity stage of harvesting than the fruits harvest at pre- and optimum maturity stage. Similarly, the TA of fruits among the cultivars varied significantly during the storage and Chaus (0 31 %) and J)a. !wwhari ().314%) 136 cultivars were recorded of 1aw t TA at 1511 day of storage (Table 4.i.5). Similar results were also repotted by (owda and R.amanjaneya (1994) and Rajagopalan (1997) fot Chausa and Dashehari cuitivar. They have reported that the acidity of fruits of these cultic varied between 0.1 % and 0.32 ° when ripened.. Furthermore, the Table 4.1.6 shows the combined effect of factor harvest stage and storage period, (H3* P) on the TA of fruits. At each harvesting stage, as the storage period increases the TA decreases signifleandy. The TA of the fruits fwd to be decreased from fresh to the 15 day of storage at pre-, optimum and ever maturity staff such as 2.741 % tia. 0,427 % (84.4 %), I,939 % tv e.359 (82.0 %) and 1.301 o to 0.270 (79.3 ?/o), respectively ('Table 4.1.6), Thus the Mirig of the present investi Lion indicated that fi-uits harvested at over maturity stage when ripened are 1e s sour in taste than ripened fails at other stage of maturity. 4.1.9. Total fiarotenolds The variations among the cultivars, harvest stages and storage periods in relation to total carotenoids are significant (p Furthermore, the Table 4.1.4, 4.I,5, and 4.1.6 show that how the interactions between harvest stage and c}tivar, storage period and dultivar, and harvest stage and storage period have affected significantly on the total carotenoids content of mango fruits. The total 137 carotenoids content was investigated highest (0.902 IOOg) for the Chausa cultivar fruits harvested at over maturity (Table 4.1.4). In addition, the Chasa cultivar fruits when ripened (at 15th day of storage) found to he highest total carotenoids content (1,404 rng/lOO (Table 4.1.5). However, from the Table 4.1.6 it is evident that the fruits harvested at optimum maturity are superior in relation to total carotenoids when ripened (at 15th day of storage). Total earotenoids •content of fruits harvested at my stage maturity, thus, increased significantly when stored for ripening and similarly, the total carotenoids content of fruits of any cultivar increased significantly as the maturity changed from pro- to optimum and then aver-maturity stage. However, the fruits harvested at optimum maturity stage and ripened after storage recorded highest total carotenoids content than the ripened &nits harvested at other stage of maturity. The present study is a close conformity to the Vazquez-Caicedo et al. (2004, 2005) who have stated that the severil biochemical changes occurs rapidly during ripening after harvest depends on cultivar, stage of maturity at 'hest and post harvest conditions among which carotenoids synthesis is one of the most important. 4.1.10. Color values There was a significant (p 1.O5) change observed in the peel color of mango fruits among the cultivars, among the fruits harvested at three maturity stages (pre-, optimum and over maturity stages) and during the storage period (Table 4.1.3). The L (lightness) and b (yellowness) values of fruits were increased significantly when harvest stage changed from pre- (45.89 and 25.95) to optimum (48.11 and 28.40) and over maturity (50.60 and 33.62), respectively, and also fruits were ripened during the storage (Table 4.1.3). Consequently, the `a' value (greenish colour) of fruits reduced significantly from pre- (-7.29) to optimum (- (i.730) and over maturity (-5.615) stage of harvesting, and during the storage (-9.342 at 0 day to -2.106 at 15" day), might be due the degradation of chlorophyll as a consequence of maturation and ripening of fruits. Although, the development in peel valour is accompanied by ultra structural changes associated with chioroplast-to-chrnmoplast transition when ripening proceeds (Siddiqui and Dhua 2009). Pulp carotenoids continue to increase in the, detached fruit as ripening proceeds (John et al., 1970) and affect the peel color. Since, the carotenoids level in the ripe or ripening fruits varying among cultivars, a significant variation among the 138 culti rs (Table 4.13) in present study is justified. In addition, the moan, grand scan, standard error in u.ean and standard error in grand mean values are given in the Table (4.1.3) Further, the combined effects of the main factors on color values `L', 'a' and `h' were also analyzed and found to be highly significant {p The findings of present investigation, thus, indicated that the fruits harvested at over maturity stage were superior to the fruits harvested at optimum maturity (partial ripened on the tree) stage in relation to the fruit's physical (weight, shape & size, colour values etc.) and biochemical properties (TSS content, total carotenoids, titratable aridity). However, the mango fruits harvested at optimum harvesting stage were also sh Combined factor Parameters Harvest Total Titrftable Total Specific L a b Stge(HS)*Storige Soluble Acidity aro1euoids gravity periods {3P) SQlids HI S1 8.100 2.741 0.104 0.945 36.287 -10.34 28.696 HIS2 11.517 2.203 0.148 0.896 42.669 -9.415 26.433 HI S3 14.167 1.322 0.342 0.81 45.709 -9.778 25.883 HIS4 15.708 0.820 0.619 0.998 48.065 -8.91 26.899 11185 17.100 0.603 0.709 1.013 49.776 -4.548 26.276 1-1186 18.925 0.427 1.007 1.036 52.819 -1.001 21.495 H2SI 9.042 1,989 0.155 0.408 41.412 -9,391 26.539 H282 12.583 1.531 0.191 0.982 46.781 -8.885 28.096 H233 14,708 0.907 0,276 0.999 47.467 -8.324 28.574 H2 S4 17,650 0.656 0A06 1.017 49.401 -7.291 27.120 H2SS 19,292 0.439 0.693 1.022 50.897 -4.460 29.496 H2S6 20.492 0.358 1.135 1.037 53.073 -2.031 30,540 H3S1 10.117 1.301 0.209 0.983 44.407 -8.295 30.217 H3 S2 12.925 4.880 0.352 1.007 48.746 -7.806 30.320 H383 14.753 0.848 0.498 1.013 50.530 -5.258 31.954 H3S4 15.450 0.544 0.662 1.022 51.531 -5.522 33.417 H385 21.225 0.395 0.782 1.034 52.509 -3.525 35,782 H386 22.667 0.270 1.080 1.057 55.864 -3.285 40.052 Sta.drd ester 0.267 0.112 0.052 0.019 1.241 0.4$8 1-.091 Significance level ** * :'" NS non signifies and * - significant at least at 5% level Hl, H2 & N3 - pre, optimum & over honest stage. 1511 S2,S3, 84, 85 & S6 - Storage period at interval of 3 days from the fresh condition 140 4.2. Non-destructive quality evaluations of mango fruits Computer vision techniques based on four different cameras (monocluorne, color, ultra- violet and ne sr-infra-red) and two different lighting regirne-(visible and CV) were used for the quality evaluation of mango fruits non-dcsfructivey. Monochrome, color and near-infra-red cameras were used to capture prof le image of fruits illuminated with visible lighting regime - while UV camera was used to capture the reflected UV image of the mango &u1t when illuminated with UV light. The acquired images were processed, analyzed and the results obtained from these camera based computer visions systems have been discussed in this section. The selected particle analysis parameters were separated in two groups. The first group covers measures of size while the second deals with various aspects of the particles' shape. Within each of these categories, there are a variety of individual parameters that can be measured from the others, which are measured directly. 4.2.1. Non-destructive quality evaluations of mango fruits using monochrome computer vision technology 4.2.1.1. Size measurements Aaaal length, width wid thickness of mango fruits The actual length (major diameter, L), width (intermediate diameter, W) and thickness (minor diameter, T) of different mango fruits treasured using digital vernier caliper with least count of 0.01 rnm weru found as 100.02 mm, 58.90 mm and 50.64 mm for Chcmsa cultivar and 1 O2.b3 mm. 57.6 mm and 52.83 mm for Dashehari cultivar, respectively. In contrast, the mean values of length, width and thickness of same group of fruits of Chausa and Dashehcri cultivar measured by image processing methods were evaluated as 97.59 mxn 57.O8 min and 491.22 ram and 99.56 mm, 56.1 7 mm and 50.23 rnm, respectively. A detailed summary of descriptive stathtica f the examined mango fruits m reported in Table (42,1 .la.). The minimum, maximum and average values of mango fruits length, width and thickness (measured with digital vernier caliper-D C and calculated with image processing (It') ` technique based on monochrome VS], and scale factors (k) for these external diameters of fruits of Ciiazisa and Dashehuri cultivars are presented in Table 4.2.1.16. 141 Table CIA Sinnnry ska tic of the m suriid physical pmprtie (]en width and thi knms) of rnang f , irk im pmr ingu~ hni cbmea ~m mcnac r mt C $. cscr;ptir~ Pbyslca1 pr pat' mea tred by Image pInR ta11~4 mum lift JJ,shehèri fruit 1tn th (inm) #ra dtIi (mu) Thi niti (rnni) Itn (rAn) B1 i (ram) ThI uuf (Imn) klep 9' ,23 Std E 0.63 0.91 x.54 x.48 Std D x,23 3I2 4.54 2,72 3.45 ~mplcudatio 394 1 0 20.61 47,16 NO 11.9 ~t;9 X1 .15 5D,42 32I5 314 5OI1 433,50 lax [ 491 624 57,0] 116,11 64,85 X6,91 Fhvc 1 prep rtiei nix urcd by t1iper (maiival) DIME60s :4Ie n error nthnrd1eIiOon sfimp vrbitIon lwmm M ¢lmuii Lii~11i. mm 100.02 0.94 6.'1U 44.1 85.1} 113.2 & L dth ([U 58,90 M5 3.92 13.34 14.52 Tbicknes~ r n 50,64 4,61 459 21,0~i (i~ltirir ligi[, mm ]~l2.0 1.0 1,7F A.19 85.46 I2A.01 1lrc , :gym 7.la 4? I55 ,51 1 ~ This , min 51.1 (M.42 ' 3.14 4717 64.13 112 ' The scale factors were obtained by rationing the measured dimetision in pixel and actual dim ion in mom of perfect rectangle of wooden shape to convert units from pixel to millimeter (mm). Complete description of algorithms used in monochrome computer vision system for Chausa and Dathehari fruits an LabV[EW software is displayed in Annexure 2a (A-2a.I & A-2a.2). • Comparison between imageprocecsmg (IF) and calif er" (manual) method The results of comparison between predicted (IP) and actual (D VC) length (a, d), 'width (b, e) and thickness (c, f) of different mango fruits of !apse and. Dashe&ari cultivars are shown in Fig 4.2.1.1, respectively. The coefficient of determination (R2) for Chausa fruits length, width and thickness were 0.991, 0.960 and 0.954 while for Dathehari fruits these (length, width and thickness) were 0.981, 0 967 and 0.950, respectively. Since, the coefficient of determination ( z) values can be interpreted as the proportion of the variance in the IP estimates attributable to the variance in the -actual measurements. Higher the R2 values, the closer the IP results are to the actual (DV) results. Thus, proposed IP method with monochrome camera set up can be yielded above 98 % accuracy in estimating mango fruits • length, 96% accuracy in estimating mango fruits width and 95 % accuracy in estimating mango fruits thickness. The second comparison was provided by means of the paired t-test results. The mean length, width and thickness difference between two methods and the standard deviations (SD) of these diameters differences for both cnitxvars are presented in Table 4,2.1.2. The obtained P-values in paired t-test analysis, for length, width and thickness, between two measurements (digital vernier caliper and IP method based on monochrome CVS) for both cultivars were less than 0.01. Therefore, the paired t-test results confirmed that the measurements computed with IP method and DYC differ significantly at the 1° level (P < 0.001) for monochrome camera based GUS. This may be caused due to the fact that the diameter of fruit is not perfectly round and hence measurements depend on the point of view. Nevertheless, the results were consistent and the correlations of fruit diameters between both methods were high. This is close conformity to the findings reported by van Eck at al. (1998) for cucumber fruits. 143 TabIe42.L1b. CO risof1 orieaur4 (YC id coputeJ (IL') ] ngL brcIth end i1li kn &munoi tni Ian in Inc analysis Fi it n rt i _ ThIdlem - swi Bulb r r O Iliv. rs VC IF la c r (k} VC IF f '! jr (k) YC IP I (k) 'h h~iji , O 83. [5 3 80 5r51 .42 1BS I3 9 ~ ,5 325 Mx 113.51 1 .9I (4, i2.O4 { $, 57,1}3 Avg, IOU 97.59 58194 57.48 w a 4922 1th° ! Mn 95 46 814 380 51.8 54,7 471 43.98 3.85 !x 12(]. a 116.11 62, 1..i ,11 S(.1 Avg. 12.63 9,76 S 11.68 56.17 52.12 22 Tab] t 4421112, P~ircd t-tcgi tjy as fix tE huit's ] igth, brii]th aid ihi±i s )sezwi il[tivRL of mange bin two iiicamirc ant ti s Inni r cal'p r am. onmpit d imp e1 lr Dimension PairLl R iündard Mean Sid t r of 95% CI far the Inr n 95% limits of Beat DevitIo db.eiic OlfOtriiice differtu a (mm) Agri ut (mm) (hih (mm) chi C18P I.math QIfQ 1 Q,S812 1.6444 R 127 1.4 :1.sl9 .0.04: . F c dth 0.tIJ 0.9 ] 1I 3 l.27 i 149:211 flRi ti 040 i], 54 ] 2121 14467 0.1697 1,11 ; 179 •,90 ]. ~JJarf i lth'ar I.en lh ?R1 1-5141 2.L31r 0.2U1l 1.7 3.34 -0,10:5.90 Hint O.8 0 0.976 L2235 Q. 1.44141 •Q14:l59 Thicness O.4G 4.95 1.57 293 0.142 1.73 , 2.47 11.3 4,80 14 In addition the measurement errors for image analysis were much lower than the once made by the caliper method. The corresponding standard deviations for major, intermediate and minor diameters differences (SD) were 0.88 mm, 1.20 mm, and 1.21 mm in case of Chausa cultivar fruits, respectively, were also found low. Slinilarly, the corresponding standard deviations for length (1.53 mm), width (0.70 mm) and thickness (1.05) mm dif rences (SD) were also found lower in cage of Dashehari fruits. Finally, the Bind-Altman plots are created for more than one measurementt m hiod comparison with v.eriiier caliper method. The Bland-Altman plot (Bland & Altman, 1986, 1999, 2007) or difference plot, is a graphical method to compare two measurements techniques. In this graphical method • the differences (or alternatively the ratios) between the two techniques are plotted against the averages of the two techniques. Horizontal lines are drawn at the mean difference, and at the limits of agreement, which are defined as the mean difference plus and minus 1.96 times the standard deviation of the differences. If the differences within mean ± 1.96 I] are not statistically important, the two methods may be used interchangeably. The plot is useful to reveal a relationship between the differences and the averages, to look for any systematic biases and to identify possible outliers. In the Fig 4.2.1.2 (Bland and Altman, 1999), the plots indicated length (a, d), width (b, e) and thickness (e, f) differences between the computed and actual measurements, for Chausa and Das ehari cultivars, respectively, were normally distributed. The, 95 % of the length differences are expected to lie between d -1.96 SD and d + 1.96 SD (known as 95 ° o limits of agreement). In additio , the outer lines indicate the 95 % limits of a ement and the center line shows the average difference. The 95 % limits of agreement for comparison of fruits length, width and thickness computed by Ip method in relation to monochrome camera set up and measured with DVC were (-0.04; 3.38). (-O50 4.20), and (-0.90 : 3.80) for CJiau,a (Fig.4..1.2 and Table 4.2.1.2) and (-0.10; 59), (-0,14 ; 2.59), and (0.30 : 4.80) for Dashchar cultivar (Fig.4.2.1.2 and Table 4.2.1.2), respectively. Similarly, the average mean differences evaluated for ChauTa and Daskehari cultivar fruits length, width and thickness were 1,64 mm, 1.83 mm, and 1.45 mm and 2.93 mm, 1.22 and 2.53 mm, respectively (Table 4.2.1.2). 145 120 y 0.91x+ 1.?? 12 'fp+5.~1 11[' ==931 ~J 9fY j f c 30 90 190 110 I?.) 80 96 W0 110 120 IN hiiagl~a h.Ian(Va) IN IffidL imn kvq a d 4 yi6]N+i2 6 . - SS J:: 74f :- i- 1 45— ' 45 3a 55 60 65 ?0 i0 5' 54 If. .5 60 6? 64 RilpyLiA.null (VC) !au ,.mu(V b e 611 R =1.95 as N_U {L, a SQ 40t -. 1 I 40 4i 50 4t) 45 50 55 60 61 MNR(. &,;5.- 1n, ) M1 emjd'EtS,vmCX) C f Fig 4.2.1.1. Comparison of computed and measured outer dimensions (L, W, T) of C) awsa (a, b, c) and Dashehari (d, e, ) fruits with image processing (IP) method based on monochrome CATS and digital vcmier caliper (PVC). 146 Monochrome camera Monochrome camera T. 3.0 u 338 a ~o 0 a # a as °p 0 2.5 a Q Q4 '{` O n . 9 QQ a V4 4 Min a P B 4 Mwn _ 2.0 1.5 —p o 4 4 44 4 1~ Q a 8 g 9 ° ° 8 O a ° a .0_I, b -1.96 Sd .... I . r I I I . 4.9- 1UO 114 E20 130 NO RS 90 95 100 L0 110 115 l;D 0 Averso Ieuj+h 1VCNS - IP] Avera~elenelttNCLH - 1P) a d 3 Monochrome camera Monochrome camera +IS8> 25 o- 2- 4 +1.96 SD 0 84 8 a ba 4 p% p 'J ~f 8 3 o- ° ° ° O ¢ G 8 O NCm O 6 O G Z 4 Q L41~Q 3.0 4 d Q 8 1.22 ~ a 18 Sao o Q 0 e B ¢ d O 4 q a 4 0.0 .L9630 0 I .014 ~' ]96Si7 rys j 43 30 52 54 36 j8 60 63 Ed 50 SS 60 rS 70 Avtrtigrx~idlh {VCM - II') Avcrga KldLb (VtM - iP} b 5 Monochrome camera Monochrome camera q -196 SD 5 +1.%SEk 4 ° 3 a a ~ fl a ~ a o 0 4 d ry~ C a 8 L 8 Q a 8 ~ a L p ° . 6 a O C 1.4 6 4 a' d .0O 4 4 4Q a - a 8 O o- - a p O ¢ 0 o- 8 4 Q O p L%sD 1 8 1 -09 • 4 1.9SU 44 45 4e 30 3x 5.4 sd 5e Go ArerntII kfli {YCM. Avemp1LLeknmi {V t - 1P} C f Fig 4.2.1.2. Bland Altman plot for the uumpariSotr of outer dimensions CL., W, T) of Char (a, b, c) and Dasheh ri (d, e. fait's computed with image processing UP) method based on monochrome CVS and measured with vernier caliper method (V M) outer lines indicate the 95% limits of agreement and center line shows the average difference. 147 From these results, it can be stated that mango fruits size has no effect on the accuracy of estimated diameters (P>0M5) and since, almost all the values obtained were within 2 SD, and so, from a statistical point of view, we can conclude that TP method is reliable. Furthermore, to describe the fruit shape attributes simple mathematical expressfons relating to the simple fruit global dimensions was used in this research. This straightforward technique has been called combination of size features by Du and Sun (2004). Among the combination of several size features measurements elongation, compactness factor and type factor obtained from the image analysis were used to differentiate the euttivars, The main drawback of this approach, however, is that the indices calculated can detect only global shape features while the loyal shape features cannot be detected by this approach. In addition, according to Saad et al. (2011) there is still some confusion in regard to shape indices names and definitions. Some of the most common simple shape indices are: (Non-) compactness or (non-) circularity (). Thus, these dimensional index (Morse, 2000) can give an idea how closely packed the shape is (not), 4.2.1.2. Estimation of shape attributes by image analysis The measurements obtained from the particle analysis of fruit's image using Lab View software, elongation, c 148 more elongated shape. The Dc-hehari fruits are more elongated in shape than the 'hausa fruits. Similarly, compactness factor (C=PIA, Where C is compactness; Pperimeter, A= area) varied between 0.65 and 0.80 alongwith standard deviation 0.04 arid CV 5.2 % for Chausa cuitivar fntits and between 0.75 and 0.82 alongwith standard deviation 0.01 and CV 1.4 % for Dasheharii cultivur fruits. As the compactness facer gives an indication of whether the lesions have protrusions outside or indentations inside of the cirwnfererxe of round or circular lesions. The value of compactness factor, thus, for Chausa fruits was found 5 0.76 shows circle indentation inside of the circumference while in case of L)as ehari the, value of compactness (?0.78) shows circle protrusions outside of the circumference. Table 4.2.1.3. The mean values, standard deviation IS.D.) and coefficient of shape geometric attributes of mango fruits ( hausa and Datheliari cultivar) analyzed by IP technique based on mouochrorneCVS. Cultivars Parameters Minimum Maximum Mean S.D. (±) C.V. (%) bausa cultivar Elongation Factor 1.88 2.3 2.10 0.0 4.3 Compactness Factor 0.65 0.80 0.75 0M4 5.2 Heywood Circularity Factor 1.04 1.25 1,07 0.03 2,5 Type Factor 0.87 0.99 0.97 0.04 3,9 Dashelwri cultivar Eiangition Factor 1.96 2.53 2.21 0.14 6.4 Compactness Factor 0.77 0.82 0.79 0.01 1,4 Heywood Circularity Factor I.05 1.11 1,Q7 0,42 1.5 Type Factor 0.98 0.998 0.995 .004 0,4 The most compact shape is circle = 4i). All -other shapes have compactness larger than 4ir. The above finding is closed conformity to the several studies have been by researchers for cstinrnIion fruit shape attributes, For instant, Wyci io et al. (2008) quantified 149 grape berry shape using commercial software and using prom they determined the area and perimeter of a grape berry and subsequently, they used these (area 8 perimeter) to calculate eccentricity and compactness values. Computer finding were compared with the data from human raters using a simple correlation. Results showed the correlations of r — 0.941 for eccentricity, and r ° 4.744 for compactness. For better results, the other measurements such as the. area, perimeter, -Max Fe°ret diameters axuj Waddel disk diameter were also obtafrod from the particle analysis of fruit's image acquired by monochrome camera using Lab View ft rare and correlated to the weight and volume of fruits. The Appendix 3 (A-3.1) displayed the tables of mean values, standard deViatiou (S.D.) and range (min, max) of some of such important geometric attributes (AMD, - GMD, aspect ratio, sphericity and area) of mango fruits (Chausa and L)ashehari cultivar in real world analyzed by IP techniques using monochrome cameras computer vision system. The area (real world. rum) displayed in the Appendix 4 (A-3.1) was calculated from the acquired image at the 0 and 1800 orientationslpostu s. Table 4.21A. Summary of coefficient of determination (R2), RSE and linear regression weight models (for Chausa and Dashehari mango cultivars) on the basis of shape attributes obtained by image processing method based on monochrome CVS Shape attributes Regression R-sgicare .-square RSE (pixel) model (Adjusted) Clairsu cultivar y 0.004x-95.78 0.875 0.873 - 11,01 ?erimeter (pixel) y = 0522x - 332.7 0.820 0.817 13.25 Max Feret's Diameter y 1.133x - 248.7 0.775 0.771 14,79 \ addel Disk Diameter y = I.894x - 372.6 0.892 0,891 1012 Dashehari cultivar Area y = 0.003x - 23.36 0.828 0.825 9.62 Perimeter y O325x - 137.5 0.798 0.795 10.42 Mare Peret's Diameter y=0.637x-61.64 0.752 0.748 11.55 Waddel Disk Diameter y 1.406x - 226.5 0.866 0 864 8.48 ISO 4.2.1.3. Relationship between fruit's shape features and weight Fig 4.2..1.3 (al, a2) shows the scatter plots of the correlations between fruit's weight and the measured area (pixel). obtained by particle analysis using LabView software, of the fruits of Chausci and Dashehuri cultivars, respectively. The area (pixel) used to correlate with weight and voIume of fruit was estimated from the fruits image acquired at the 00 orientation posture. iimple regression equation denotes the Iine that-best. fits-the-plot-of all-pairs- (area, weights of the observations obtained by ]P method for both cultivars. From the simple wear regression analysis, thus, different equations were generated by using observations and presented in Table 4.2.1.4 for both cultivars to predict the weight of flint from surface area (pixel), perimeter, Max Feret's diameters and Waddel Disk diameters. Among these, Waddel Disk diameter [Fig 4.2.1.E (dt, d2)] had highest R value i.e 0.892 for chtiasa cultivar and 0.866 for Dash hari followed by surface area of fruits i.e. 0.875 for Chausa and 0.8286 for Dashehari cultivar. It suggests that more than 89 % and 86 % variation in the data (weight), in case of Chcwsa and Dashehori cultivars, can be explained by the linear regression equations based on Waddel Disk diameter and more than 87 % and 82 % variation in the weight of fruits of Chausa and Dashehari cultivars can be explained on the basis of area, respectively. Fruit perimeter obtained by particle analysis of images captured by monochrome camera setup and analyzed using LabView software is another important shape feature of the object like mango fruit. In this current research, it was taken as independent variable and on the basis of which fruit weight was estimated and found that the correlation between fruit perimeter and weight was reasonable. Figure 4.2.1.3 (bi, b2) showed the scatter plots for the relationship between fruit weight and fruit perimeter (pixel) of Chausa and Dahc sari cultivars, respectively. The R2 value (0.82) for Chausa cuttivar was found noticeable higher than for the Dashehari i.e. 0,798. Thus, proposed IF method with monochrome camera set up can be yielded above 79.8 % accuracy in estimating mango fruits weight on the basis fruit perimeter. In addition, some measurements dice Feret's diameters, the major axis, and the minor axis (Zheng et al., 2006b), have been developed by researchers and are used in the food industry, to avoid complication in measurement of length and width of irregular shape objects. 151 Max Feret diameter [Fig 4.2.1.3 (c1, c)1, in this research, has also been correlated with the weight of fruits of Ghui and Dashehari cultivars and found that MaxFere 's diameter (R2 value; 0.77, 0.75) were slightly loosely cnrr sated with the fits mass than the Waddel Disk diameter (lea value; 0.89, 0.86) of both cultivars. The values of intercept (from -95.78 to -372 and -23.36 to -226,5 in case of Chazisa and Dashehari cultivars, respectively) are of no practical predictive value. Similarly, low regression coefficient values (from 0.004 to 1.89 and 0.003 to 1.40 in case of Chausa and Dashehari cultivars, respectively) are also have less important and fitted value for regression coefficients no longer tells us the effect of a 1-unit change in X (area, perimeter} MaxFeret and Waddel Disk diaDleter; pixel) of Y (weight; gm), In spite of above, the measurements of area and perimeter are more preferable to the length and width measurements for the evaluation of the size/shape and to develop a relationship with the weight and volume of products such as fruits and vegetables. No matter how irregular the shape of the object, or what its orientation is, measurements of area and perimeter are stable and efficient once the object has been successfully segmented from the background (Zheng and Sun 2008a, b). 4.2.x.4. Relationship between fruit's shape features and volume A simple linear regression analysis of fruits volume and measured area (pixel) is shown in Fig. 4.2.1,4 (a,, aa) for Chausa and Dashekari cultivar, respectively. The value calculated for Rz were 0.85, 0.79 for Cha.wsa and Dashehari cultivar, respectively. These values suggest that more than 85 % variability in the data (volume) of Chausa fruits and more 0.79 % variability in the data (volume) of Dashehari fruits could be explained by the linear regression. Similarly, the fruit perimeters [Fi . 4.2.1.4 (at, a2)] were also correlated to the volume and the R2 values for cultivars, Chausa and Dash ari, were evaluated as 0.765 and 0.778 [Table 4.2.1.5], respectively. From the Table 41.1.5, it is evident that the fruit area (pixels) has higher correlation with fruits volume than the fruit perimeters. Furthennore, the Max Feret's and Waddel Disk diameters were also correlated with the fruits volume and the observations were summarized and displayed in Table 4.2.1.5. 132 ='0 Y 0.OWls-95.0`9 50 y=OQO31x•23.305 a3 0919 SOU l =QS?5? 00 150 l5b r 20U IUQ CL 3o rt 20000 -000C. 60000 SGh0h 100 H 0 aaa 500Q 60NO 70CM 00'D &KXP Area (pi dl ) Area (pixels} a1 a2 350 =0.5114,E•332_.? 250 y= 0.3255x-13T$i R"= 0.8201 ~=R.79R4 A =04 2C4 .lIIi,I, 150 la•a 1{0 30 J0 fi0C 00 0 900 10{X Liao 570 7C0 9(}0 1100 1300 PauuekeroffiiiLL(pixe[ ) Paimslaorbils ixdt) 1.133+,;-'_4872 150 i R"--0.753 r3l p AlLilti I: ° 100 100 ~°— is 200 10 344 3 4U0 40 2O 250 300 350 -0 450 Al %Rddimie~er (pie]) M Feret Ji t r ( g) Cl C2 9U 4 y 1.40 • 2!6.51 p=1, 45 -3716; • R00.3669 go ' i{ 18 ' ?04 = 1S0 - U~ 150 q 100- + 1VV 50- ? 0 50 I ?OQ 'S0 349 35 I, ?C0 250 300 350 d1 d Fig 4.2.1.3. Coffelations between s]iape attributes [area (a1, a2), perimeter (b], b2), Max Feret's (c', C2) and Waddel Disk (di, d2) diameters] of Chausa (1) and Dashe ari (2) fruits computed with particle analysis by image processing (lIP) method based on monochrome CVS and weight of fruits measured by digital electronic balance (DEB). 153 Table 4.2.1.5. Summary of coefficient of determination (R2), RSE and linear regression vohnne models (fir Chausa and Dasiehari mango dultivars) On tho basis of shape attributes obtained by image processing based on monochrome CVS. Shape attributes Regression square R-square - RSE (pixel) model (Adjusted) Chausa cultivar -- - Area y=0.004x-91.E1 0.858 0.855 11.40 Perimeter (pixel) y = 0.488x - 306,7 0.765 0.761 14.66 Max Ferret's Diameter p I.053x - 225.5 0.713 0.708 16.20 Waddel Disk Diameter y=1.820x - 357.7 0.877 0.875 10.56 Thshehari cultivar Area _0.06 x-16.Ef 0.792 0.788 10,19 Perimeter y = 0.309x - 126.6 0.778 0.774 10.51 Max Feret's Diameter y = 0.606x - 54.68 0.734 0.729 11.50 Waddel Disk Diameter y=1.336x- 1L4 0.846 0.843 8.75 Like fruit weight, the voIume of fruit was aIso found to be highly correlated with Waddel Disk diameter of fruits than the other parameters such as area, perimeter and Max Feret's diameter. Fig 4.2.1.4 (ci, ci) and (d,, d2), show the scatter plots of relationship between Max Feret's diameter and Waddel Disk diameter of fruits and volume of fruits, respectively. The R2 values, >0.71 for Niax Feret's diameter and 0.84 for Waddel Disk diameter, showed reasonable and good fit of the data, respectively. 4.2.1.5. Geometrical attribute based modeling of fruit weight and volume Modeling off man o ruits lna s For the fzuits mass calculation, 20 models (caIibotior and validation) were suggested based on geometrical characteristics (Table la). Model no. 13, 14, 15, 16, 1$ and 19 (in Table 1a) were had the highest R value about 0.93 (for both calibration and validation) and lowest SEC value between 09.09 and 09.25 and SEP value }between 09.44 and 09,67, For calibration models, the R value was varied between 0.671 and 0.934 while for validation models; it was varied between 0.651 and 0.928. 154 21 r —00O2- ..17 250 1 — U O ls- 1i1d • 1 4-7 l 095$ O0 1 ?00 150- o l00 i 100 2/y /~ 14 50 V 40000 5050D 6000 10000 80060 9(KIO0 -NU 40010 fflG0 .]000 i0000 ,iM riEels) al a2 y = 4,369? -12&64 .. ;=01888x-306.7' R D. rS2 , ° x ZQD I 150 150 a C ILO` 100 5Q ~ r 610 700 $ IKI 9O0 10 0 0 1100 5 0 °04 900 1100 130 Fftirsxet rof$oil {pi X1:1 P&mek• tffiuit (aixls) bl ,~ 2 1 Y=1.053&"-??553 50 - T=04M1X•34,(4 c E 11 R'r 0.71-} ,04 R'=9-'318 I 4M r~~' -: R 1110 a r I I 200 '5(1 300 350 - 10 450 2410 250 300 310 40 450 bi~xFa-et di~:zu~tee (pixels) M Fck dnutt4 (pixdh) Ct C2 '5Q -1.3369.- 211-45 ' g'=0.3778 R"= 0•S-Ibfi ?00 I U 1a 100 50 ,- 2041 '5{' 340 350 i1 0 300 W [dddi:'~c&m tc(p rL0 IYAdeldiA-4itre[er (pixels] dl d2 Fig 4.2.1.4. Corrn1atioriS between shape attributes [area a az], perimeter (b1, b2), Max Feret's (cl, c2) and Waddel Disk (d1, d2) diameters] of Chausa (1) and T3askehuri (2) fniits computed with particle analysis by image processing (JP) method based on monochrome CVS and volume of fruits measured by water displacement method 155 Similarly, the standard error in calibration/standard error in prediction was varied between 9.0919.44 and 17.99118.43 while the difference between SEC and SEP, which help in selecting best model, is varied 0.26 (model no 4 & S) to 0.51 (model no I9). The model no, 19, however, had a bit (--0.001) lower power for weight prediction than the model no. 13 and 14. Further, among the models 1, 2 and 3 based an single variable model no 1 based on fruit's length had higher R value and lower SEC/SEP for both calibration and validation results with no significant bias between them. The R value of the calibration and validation model based on IP length of mango was found 0.870 and 0.863, respectively, The R value of the calibration and validation model based on iP width and thickness can also be noted from the table. Model 3 among the models 1,2 and 3 had the lowest power (1R value-0.67 I) of mass prediction which differed from the findings by Tabatabacefar and Ra1ipour (2005) recommended a model to calculate apple mass, and Khashnam et al. (2007) recommended an equation to calculate pomegranate mass on the basis of minor diameter. This difference might be due to the result of the dissimilarity between the shape of fruit (i.e. between a mango and an apple or a p9me anate). Similarly, the model no. 7 azid 8 were based on geometric and arithmetic mean diameters have higher coefficient of correlation Le, 0.819 and 0.916 for calibration, and 0.850 and 0.912 for validation results. But, all three diameters (length, width and thickness) must be measured for model 7 and S, which make the sizing mechanism more expensive (Soltani 2010) and complex. Modeling ofmango fruits volume Table 4.2.1.5 displayed 20 models formed by using different variables for fruit volume calculations and found that the R value (calibraticrilvalidation) of these models varied between 0.67810.650 and 0.510/0.798, respectively. The model no. 1 Teased an IP length (L) was found best with lowest difference in standard error in calibration and standard error in prediction i.e. 0.38, The calibration and validation results show that the SEC (15.62), SEP (16.00) and correlation coefficients R (0.810 and 0.798) are almost similar with no significant bias (0.036) betweon them, In addition to model no. 1, model no. 7 and 8 can also considered as best model on the basis of higher value of coefficient of correlation (R) and these models are based on single derivative variables i.e. geometrical (Dg) or arithmetic (Da) mean diameter. 156 Toble 4.2,16 Varian MR ui,deCs bisO oa IP Jimerri na'I haiicfertii (1 h m of mEng bl of ~ts dDiiv irt No. +,aduls 11 luii o ,' 'umbug c8~thr t[on V 1idLLt1 (IjbrllhiL V811 a&n C ibiatiii1 \/ Ji 4~141L i . • T r - . - - - -.- - ...- 1:3 (4L-!1 1251 12.~l O,kYJ 0IU 1 O,~NO O. ? • ~ ' I # ? * ' 1 35 167 I6.O UI ~ 8 0, 761 1).746 14 0.0) 4021 Uó71 0,6j1 4 N1=x,LV+kI i=IM104Ll-36'2 122 11A -QM U JU 4.9!3 0.8)9 I1. ' 0,12 ,C . 27 Qr 0,85] { • 1C, 'I'+, 3 1,t.4,Rnkip-1f, 4S Si 131$ Ai) U. 2 Or 61 i 54 H =1i 1? + ; a3/Jx - - 12.1 1 ,3 D.p~U ,M38 4.859 0.950 8 MRK, ,-Kj 'M06.~P -239.92 1OI1 I OP 4428 0,16 OI9 2 9 1 = ',J,4K,1#+J, 1~~=2,(Af,+1fi~ik'-'L2f S l~1.~3 W5 ],11Qij ,OI ~, 1 G.98 i0 h{=h'~~~+K,'T+K. 1`1=2,80.E+]ii7T-19(1,43 O4.~3 10.x$ ,UO1), 4, 31 x.92O 4.414 11 M=X,1,•1.92LVs .J, M-],1?L-2,g~10-1 LT-63,40 ] 25 10,68 dl.UUO -01019 0,915 O4~~ ]? M-KL+SILT+K M-1.151+1,93 i0„1T-i9.I O.9O 11),25 4.QOO 4.027 0.31 a1 13 kI-KIL#XT-h~ M=22-F201O PT-151H p ,Q (944 4,U4 -(i43' 6,934 11 14 M=,cif,+ 31r • +9 1=1 + Z i4'- T- U1~1 DORY 09.44 -IJ.u(JO -4I U. 4 0,928 15 Mi= ,1,rK.iD +R, N1=1191-~15b,-?3~~1b 4~,j1 4 5~ OAii •6 OI91 I 16 1 M=l id+rC~ 'f M=I,1H!t4 7L). MM Y.57 09.17 U9•S3 0.I -0.051 0.912 0.911 l7 9 =91 4-9+4R 4 K3 M=4,l1L+2, '1Am -~?5i 10.71 Iil1J I1 1J,14 0 6 1 ~I=h'~Ltll~sp + rj ir1-47l,t3,3~S~,. ~96 O9,~S Q9 U,QN -0.051 4.931 4. 1 I1~ ~I+1 ~4IQ~4 -0.025 4, 33 1 G1k11k'+J { +JCa M=-A.4.630 L '1' , 7 10,A 10,71 IUI U O.ILL14 C. Q7 157 'iibk 4.2.1,7, \"irJaui MLR mmk1 bescd av LP dimmi rn1 aura,6tadiatiCB 1cg th, L: brraddi, \V aid tb( I of n1ug 1is f chasm end Da i wr old' is S, Hogs RL,[Aim SECS p Digs vfr~ pion No. C liblatioR Wd~ti n if ti,ln ValluJtm Cajjt* Va1idasioR I =9~1+ 1 V_S.2 L—L42.. 15. ]6.00 b.OU') .U3i Y—K,1I'J.~# f —S,4~il~—1 ~, L 19,24 19.17 -0.040 J'.K~T I J ; V39 'T 21.77 J&,4 1$$~ 164$ 1I1~ Vol 0,M x.713 0.619 1715 U O4 0175 745 V=K,L 'T+12 V- ,80 1I rfT•I•45 4 11.2 11. I7I Or7'I1 7 V — KID FK. V -5~ DE , 2.I1 17i J71 f1 OU )IO.4 4,764 0.751 R I'=K Lid +),'} 1+= 13~1f — X4,4 16.64 ]7.7 O.Oi} 0.059 06776 1,761 K1 LtK2,Wt '3 1$.13 19.15 (I.(OU 467]8 ~ 8 1 Gf JC1,fKT4J 1741 UrU4 0,0 11749 ] 1 Gf lJ+ {+ , y=1.17L=Z,63x:D LF—a6u 18.46 1929, .4.0 U,011 0.713 0,42 12 G'=1 tL+J?+K} 1+—],PAZ+1,93X10- Ll'-99,18 l,.93 17.5 •U.UCfJ li I3 143 13 , 1 + ~1f7'tK~ v— 2,621,+ 2,61Ki1: —1MA 16,54 1731 -lf.0 1LUI U~717 ~J4 14 V=KtL+K3Li1'7'+ ', ~=I.BF~.E N10LWT-52.14 16.B 1755 '1]rO'1]r 1]iOI 11370 1741 15 /=KIL++X~ V=1,~9l+3.329 -12116 1 .4$ 1I - ~!X'O 0,7?7 0,754 16 yLK11+9~ o +Ka V-1,I L .27 y - 229.5, .6.59 1?,S 0.000 G.24 .fl7 O.iS4 17 Y -K,1.-K jAF. +1C .42 1925 715 18 V—] I+Kr$ +l v_ L-11,37S~-442. 15.55 17,31 O. n[I 1}L101 19 v0 ,t 1 X V+ ,f•l i t V=2.#)L I I.6W+1111'• 23.I)7 .6.44 17.44 0.4~U 4.446 0719 -50 Y= i I l '+, } Vm. 184 l9I r , f10 E 1 40 0,7E 0.631 However, all three diameters (e gth, width and thickness) are required to measure for model no. 7 and 8. In this case, thus, the other models with multiple variables but have higher R value and louder value of E /SEF can also be used. Similar findings have also been reported by Rehkugler and Throop (1989) and Thioop et al. (1995). They have attempted monochrome camera based computer vision system for on-line grading of apples in their research. According to them, the monochrome camera is _widely used, so far, in computer vision systems as color camera to classify the products (mainly external qualities) based on their surface phenomena such as size, shape, color, texture, and other visible characteristics. So, it can be concluded that the monochrome computer vision can be used not only for sorting and grading of mango based size but also can be used for the estimation fruits weight and volume by modeling technique. 4.2.1.6. Detection of external defects of mange fruits using monochrome computer visiGn technique Various images of mango fruits of Chausa and Dashehari cultivars alongwith several defects such as, mechanical damage, shriveled, scratch turns blackish and defected and spoiled with nnthracnose disease were used to test the developed algorithm (Fig. 3.19). The defects on the fruit's surface were segmented by applying the thresholding method. The pixels were partitioned in this method depending on their intensity value. Among the various global and local thresholding methods, the Iocal thresholding: background correction has been applied for detecting the defects. First, background correction performs to eliminate non-uniform lighting effects and then performs threshotding using the interclass variance thresholding algorithm (Otsu 1979). The developed algorithm was found able to detect defects on the surface appears dark in color. The black decay spots on the fruits surface caused by anthracnose and mechanical damage after harvesting was segmented efficiently using the developed algorithm. The scratch on fruits surface which turned dark after harvesting during storage can also be segmented. Fig 4.2.1.5 shows monochrome images of severally damage fzuits (10-49 %) because of anthracnose and related defects after segmentation. The black spots on the surface of fruits 159 monochrome image are presented by white spots after segmentarion on the dark background. Similarly, Fig 4.2.1.6 also shows the images of various slight diseased/defected fruits of both cultivars with defected area betvre n I and 9%. In addition, Fig 4.2.1.7 and Fig 4.2.1.8 are presenting the images of mechanical damage and shriveled fruits with various defects on the surface, respectively. The relevant infected arias were also showed. below in the image as white spots after segmentation,. The above findings are an agreement to the results reported by Rigney et al. (1996) for detecting sears, cracks, and spreading tips for asparagus using monochrome computer vision technology. In addition, the present finding is a close conformity to the finding reported by Davenel at a1. (1988), Rehkugler and Throop (1989), Singh and Delwiche (1994) and Throop et al. (1995) for the detection of blemishes and braises on the surface of apples. The area, diameter of all white spots (defects) and number of defects can be evaluated using particle analysis (included in Lab View software). Consequently, it can, therefore, be concluded whether the fruits is fresh or spoiled or severe defected or slight defected on the basis of percent defects and its sip obtained by particle analysis. This is an agreement to the scales repotted by Smoot and Segall (1963) and Koomen and Jeffries (1993) that the disease severity can be defined mainly by either size (diameter) or number of spats, respectively. Some fruits were found with many small spots or some others with a single large lesion which created problems during the assessing the severity of disease accurately. The area measurements of the spots were found crucial in this case. For instance, Brodrick's (1978) given a scale calibrated as a function of a subjective estimation of the percentage of the surface affected. Thus, calculation of percent defected area on the fruits surface can also be helpful for defining bruise severity. The defected areas on the shriveled fruits can also be detected by this algorithm (Fig 4.2.1.8). The white spots (4, 5, 6 and 10, 1!, 12 in Fig. 4.2.1.8) on the dark background are the respective segmented defect area on fruits surface (1, 2, 3 and 7, 8, 9). The Fig. 4.2.1 ,8 showed no evidence of shriveling condition of fruits after segmentation. However, wrinkled surface in the image of shriveled fruits captured by monochrome camera is seen. 160 Fig 4.2.1,5. Images of severally damaged (10-49 %) fruits with anthracnose disease captured by monochrome camera and of related defects an their surface after segmentation. 161 Cif 42. i, ML LIJI I QILL iuk c of slight fiuiis W ] fit tci anna 1- f (1 tw 4 any 9 to 12) acrd im ac afar hffii se~ttt tatiem (5 r B ind 13 to 16) oi (heir 5ti: tee, 112 1 2 3 4 5 6 Fig 4.2.1.7. Monochrome images of mechanically damaged fruits (1, 2 & 3) and segmcnte.d defects (4, 5 & 6) on their surface. Fig 4.2.1.8. Images of shaivelled fruits with various damages (1, 2, 3 and 7, 8, 9) captured by monochrome camera and segmented defects (4, 5, 6 and 10, 11, 12) on their surface. 163 Finally, the performance of developed algorithms was also tested on the basis of its accuracy and efficiency in differentiating the bruised mangoes fruits from the unbruised. Both ac. uracy and efficiency of algorithms found to be decreased as the severity of defect increased. The maximum accuracy was observed for fresh or slight defected fruits and minimum for the severely defected fruit. The overall accuracy of algorithm was found to be 88.75 ° and the efficiency processed by this technique was found to be 97.88 % (Table 4.11.8). The average inspection time taken by algorithm was lowest for fresh fruits i.e. 9.69 rns. However, as severity of defect increases, the average time for inspection also increases and the very severely defected fruits have taken longest average inspection time i.e. 31.77 ins. Among the various algorithm steps, the color plane extraction takes longer time to perform the inspection -on the images, The main benefit of this algorithm is its ability to inspect images captured even under non-uniform iliwrnination in very short time (ms). But, the main drawback of this developed algorithm is its inability to detect the bruise near the edge of fruit. However, the bruised arm away from the boundary are detected accurately. In spite above, monochrome camera based computer vision requires less overall inspection and image processing time and thus it is more suitable for on-line tools. Since, the major challenges for on-line inspection are to produce quality images that provide clearly identifiable features and to have both efficient hardware and software to process the images fast enough for on-Line implementation. The monochrome camera is relatively cheaper compared to infra-red and other cameras. Table 4.2.1.8. Performance of proposed algorithm Parameter Minor Severe Very severe Fresh or slight defected defected defected defected Accuracy (° 6) 91 90 78 96 Efficiency(%) 98.5 97.2 96.01 99.8 Average inspection time (ms) 10.34±0.19 1336±0.10 31.77±2.36 9.06±0.14 No of fruits 25 25 25 25 Defects (%) 1-9 10-49 49-100 0-1 164 4.2.2. Non-destructive quality evaluations of mango fruits using color (RGB) computer vision technology 4.2.2.1. Size measurements AeJual lengt, width atul thickness ofniangU fruits Table 4.2.2.1a displayed the summary statistics of the outer diameters (length, width and thickness) of mango fruit's measured manually using the caliper as well as measured by image processing method based color computer vision system. From the Table 4. .2.1a, it is evident that the descriptive statistics such as mean, minimum and maximum of outer diameters of mango fruits of both cultivars measured by former method (when color camera set up was used for image acquisition) were found slightly higher than the values obtained by latter one. But, the descriptive statistics such as standard euor, standard deviation and sample variation were lower in case of image processing method than caliper measurement. Table 4.2.2.1b also presented comparison of measured (VC) and computed (IP) physical properties of mango fruits in temps of the minimum, ma imurn and average values of mango fruits external diameters and scale factors (k) for length, width aad thickness of fruits of Chausa and has/ehari dultivars. The complete algorithms description on LabVIIEW software used in color computer vision system for outer diameter measurement and extraction of morphological features of Chaze.ca and Dashehari fruits are displayed in Appendix 2b (A-21D.1 &, A-2b.2) separately. Comparison between image processing (IF) and caliper (manual) method Fig. 4.2.2.1 shows the comparison results between predicted (IP) and actual (DVC) length (a, d), width (b, e) and thickness (c, f) of fruits of Chausa and Dashehari cultivar, respectively. The coefficients of deterui natkon (R2) for Chausa and Dashehari cultivars fruit s length, width and thickness were 0.984, 0,935, and 0.925, and 0.979, 0.937, and 0.929, respectively when color (RGB) computer vision system was used. The proposed IP method in this case has yielded above 97.9% accuracy in estimating mango fruits length, 93.5 % accuracy in estimating mango fruits width and 92.5% accuracy in estimating mango fruits thickness of both cultivars. Since, the coefficient of determination C 2) values can be coiisidcred as the proportion of the variance in the IP estimates attributable to the variance in the actual measurements. 165 TaMI UJ-Ii, Su~+r~ar'sk his ci of is sic; i 1 hv'i al 'rap r~i s (Ie h WridtA aid Cbi k s) if m3n,go ftuit'a uaiii clipff mli?m~t~ i bie1 Unpb1ort'V , Delplive PIiyskAI propchlks measured by Ini a pron _ . -- • _ _ EtAtics ri+,g fruit gIar fruit L.rOh (mm) $rea&h (u ]i i Th ±nc- (mm) L Ih {i Bmdtf1'i[ ]') 11i. I~ Gen .7) 56,15 47. 5557 . 5.46 SWF O.V ~- .4 O.P W 0.35 S I) 6i1 x,14 4,18 6,29 2,37 2I6 Bump 9.59 DA 39~~') 3.62 . Z iii 2I1 19,7 3293 241 .61 46,24 Mai 112.91 6152 57I9 1111,14 62.45 5539 ?hyckWW p mp ti c m!a ur&d by CaIiptr (manual) Min &iridara crrUr S1drd D YMOD Sa1eyzm' t1 a MIi iaiim MaiiuhlIll__ - Crr rultivar 1cngth, mm 10 '194 , 44 I .Rl tii3Y1 3r11h,mm 5 (1, 1 154 64,98 Thi essa tri 50.64 4.64 4.59 21IJG 33.9 M$.69 D~rgi rr fivar I ltl arm 102 I 7.7S 119 12.I rdth,mm 57.65 +).47 Z55 6.51 5L'i •!hkhu5T I1]m 52.83 x.42 3.14 9I84 47.27 _x.13 166 Table 4111b. C mpi d uii odmrm d (VC and iJnmpwcd (IF) 1 u11 br uith and cl ,tess of go flu[rb u5 1 iu Chc analy~ia Fruit D ri~iv Lu th Scale Bath Sik TIitthiesi $eat? c~Iti4ars e il~lliti YC IP f4bu1' 1k YC IP Tatar (k}. VC IP fctiw (k) 5U15248;75 3,F_ 33.5 32 3 3~ Sfax k ~ I , 1 b , $ 6L52 M~.R 5159 Avg, 1001 79 5 % 56,15 I.64 475 D ~':r lw l 85.6 $x 41 1$B 517 5'L7 3,8i 47 2 4'14 .0 120 01 11f.14 0.93 10.4 60D Avg. 1021 7,3 5768 515,7 32J2 5c 4 T~bOO U11:i: 1iir1 1-test D:~a[M, fr fr 1.1eng',1 &h aad thk:kii a k WOM two I uxmcnt mcl and [vernier colipa 5 tiiipi1ied by inug uciiiiu [Ii cnion Id 1 Standard Mean S#d amr Df 9% CI fir the end 9S%Gmiftof t.trst D Iasi n different Owe difference (mm) Akrimui (mnO) (mm] (mfl chow tullIY]r lr:glh 1}.0D O.84 1.3105 A5 271 3.4$ 0.5'): 5r flrdth 0iJ O.L)35 l.494 2133 : 3.17 NN : Ihiobm O.Srf~ 0. 1.5395 2, ril 0 , it 56 ~,IJS 3 'U.~O b.GU Das irk cltl 'ar Lflgth OX 0,9? 1.572 1 4 ~ 14 ii,LJ Bud$ b, O.7 0.$680 1.5936 4.1150 1.36 ;1.$1 :1.19 Th i.9 1. ] o 13611 1570 59 ,, 6I63 th , 4.7 The higher the R2 value, the closer the IF result to the actual. (DVC) result. The Chausa fruits length which have higher accuracy as 98.4% if estimated would be the closer to actual results. On the other hand, the results of paired t-test revealed that the length (major diameter), width (intermediate diameter) and thickness (minor diameter) computed with image processing method and DVC were statistically significant at the 5% level (P < 0.05). The mean length, width, and thickness differences between IF and caliper method are presented in Table 4.2.2.2. The mean difference between fruit's length measured by caliper methods and image processing was found highest i.e, 5.20 for Dashehari fruits. The obtained P-values of fruits were found less than 0.01 both for cultivars and the standard deviation (.D) varied from 1.31 to 1.54 and 0.87 to 1.57 for Chazrsa and Dashehari cultivars, respectively. Finally, Bland-Altman plots are created for more than one measurement method comparison with vernier caliper method. The Bland-Altman pint (Bland & Altman, 1986, 1999, 2007), or difference plot, is a graphical method to compare two rneasuremenns techniques. In this graphical method the differences (or alternatively the ratios) between the two techniques are plotted against the averages of the two techniques. Horizontal lines are drawn at the mean difference, and at the limits of agreement, which are defined as the mean difference plus and minus 1.96 times the standard deviation of the differences. if the differences within mean f 1.96 SD are not statistically important, the two methods may be used interchangeably. The Bland-Altman plots, Fig 4.2.2.2, indicated length (a, d), width (b, e) and thickness (c, t) differences between the computed and actual measurements were found normally distributed for both cultivars (hausu, Dasliekari), respectively. The 95 % limits of agreement for comparison of size (length, width and thickness) computed by YP method and measured with digital vernier caliper (DVC) were (0.50; 5.70), (-0.20; 5.70), and (-O.UQ; 6.00) for Chaiisa fruits [Fig. 4.2.2.2 and Table 4.2.2.2.). Similarly, for Das rehari cultivar the 95 % limits of agreement were obtained as (2.10; 8.30), (-0.11; 3.29), and (p.00; 4.7), respectively (Fig. 4.2.2.2 and Table), From these results, it can be stated that mango fruits diameters has no effect on the accuracy of estimated size (P>0.05). 1sS •-• 11 ' = .93S'+ 1.152 R=-,93 1 310 -I 110 A==U979 rt 100 = 1+}0 1' I 60 I 1:-i So 40 10 110 120 511 10f1 1I{) 120 1.111 A c1vn. Anus{VC)h liareo!a~ 71nnh(C) d 65 y=0_•39x+1 61 65 y0.866x+613+ tl R'=O 31 ~5 rl 1 43 '(5 50 55 60 65 70 51.1 52 51 56 SS 60 62 64 66 M1I1TL&1I1I.lLJJ1('iC) - - kf 0wid1,ii n b 60 y0788xt4 331 :J, A$1 k o 4S 40 . 35 10 45 50 55 60 65 Mc'thuiknulill (VC Mango LhicE . iuii (VC) C f Fig 4.2.2.1. Comparison of computed and measured outer dimensions (L, W, T) of C1iaucc (a, b, c) and Dasi+ehari (d, e, f) fruits with image processing (II') method based an color CVS and digital veraier caliper (DVC). 169 Colour camera . +I• _ Colour camera +x_96 sn n a 4 O W' a 5.7 6° 8 8 a +~ O 6 a 8 a 8 4 . d a ~p 4l~ O Y, 3 4 3,1 q° 6 a p0 p d q 4 4 a 6 O ° -1cI62d h -1,45 so 2J 0 0.5 a 11p 120 130 9 9p 95 100 [45 Ito [] 00 1L t4verESPImpdh [YCd1 - 1P} A~exegalcngih {YQM - 1P) a 4.0 Colour Camera Colour camera 35 +I.00 sn +1.96 5D •' 2 0 ~a 4 a,7 2.5 ° ¢ 4 Y a Q 2.0 o a° - 4 ° a 4° Jw can 4 9 ML l. Q 1..~ a 2.L9 2 O 4 do g° ~,O ° 4 +~ /Y a O ~} 1 0 d84 8 F ~ d O a d a O • •1.965D -0.][ -0-2 S 50 52 54 5b SR 6d 62 64 6d 46 53 SB 63 f~A A►~rage »Idt1 (1+Gd1 - ]P1 Avcrago »1dlh tVCM - IP} b 6 Colour camera Colour camera 6 +[36SI7 - 1 SD 6.0 O ac. 4.7 $ b 4 0 ~y # ° O •0a 4 ° a 3 o a 3 a Mean 4 a ° 0 3.0 2 oa 4 0 ° a 2.4 2 o 8 4 p a 1 O 4 4 4 bap a ° ° ° ° -1.06 SD .4d Sd 0 0.0 -0.4 -1 40 45 SO 5S 60 46 .l 54 52 54 36 5K 64 Avac ILicknr,s (VCM . IF) Average thkI i (YQMI - 1F'J Fig 4.2.2.2. Bland—Altman plot for the comparison of outer dimensions (I-, W, T) of Ghaysa (a, b, c) and Dashekari (d, e, f) fruit's computed with image processing SIP) method based on color C.VS and measured with vernier caliper method ( CM); outer lines indicate the 45% limits of agreement and center line shows the average difference. 170 4.22.2. Estimation of shape attributes by image analysis Table 4,2.2.3 displayed sip dependent shape attributes of CIwua and Dashehari mango fruits. Compactness, among them, provides a good example of how to describe shape. The mean value of compactness of CJaiws fruit was 4.7 0.03 (CV, 4.1%) and of Da hehaH f nit uras 0.78*0.03 (CV, 4.2) which were less than 1 i.e. largest value of compactness of a perfectly circular food product. Variation in compactness shows variations of the shape of fruits. As more and more corners are added to the product, gradually reduce the value of compactness. Similarly, elongation factor obtained by image processing, as in case of monochrome CIS, can be used to differentiate Chcwsa and Da. hehari cultivars and found that the Dashehari fruits are more elongated in shape than the Chausi fruits. However, Heywood Circularity and Type Factors of Chausa (1.07±0.04, 0.1i),03) and Dashehari (1.08E0.03, 0.99±0.03) cuItivars were found almost same. Differentiation between the cultivars on the basis of these factors (Heywood Circularity and Type Factors) would be difficult a little. Table 4.2.2.3. The, mean values, standard deviation (S.D.) and coefficient of shape geometric attributes of mango fruits (Chausa and Dashehari cuItivar) analyzed by IP technique based on color CVS. Cukivars Parameters Minimum Maximum Mean S.D. (±) C.V. (°lo-) Chausa eultirar Elongation Factor 1.96 2.33 2.11 0.09 4.4 Compactness Factor 0.66 0.80 0.76 0.03 4.1 Heywood Circularity Factor 1,04 1,34 1.07 0.04 3.8 Type Factor O.3 0.99 0.98 0.03 2.9 Dashehari cultivar Elongntion Factor 2.02 2.57 2,23 0.15 6.7 Compactness Factor 0.64 0.83 0.78 0.03 4.2 Heywood Circularity Factor 1.05 1.22 1.08 0.03 2.7 Type Factor 0.85 0.998 0.99 0.03 3.0 171 41242.3. Relationship between fruit's shape features and weight The simple regression models for estimation of weight of Chausa (a) and Dashehari (b) fruits on the basis of fruit's surface area (pixels), perimeter, Max Feret's diameter and Waddet Disk diameter obtained by image praising have been indicated in the Table 4.2.2.4 and the scatter plot have been presented in Fig 4.2.2.3 (al, az'); (b1, b2); (el, c) and (di,di), respectively. The area (pixel) used to correlate with weight and volume of fruit was estimated fi m the fruits image acquired at the 00 orientation! posture. The coefficient of determination and regression standard error of estimate, presented in the Table 4.2.2.4, denoted the lines that best fitted to the plot of pairs (area, weight) and (WDD, weight) of observations obtained from color (R_ B) computer vision system for Chao. a and dahsehari cultivar, respectively. Higher R-square (0.765 to 0.842) and adjusted R-square (0.761 to 0.842) values Were evaluated for Chaua cultivar followed by the Dashehari cnitivar i.e. 0.732 to .810 and 0.727 to 0.806, respectively. However, the lower RSE was recorded for Dash a. cultivar (10.12 to 12.02) than the Chusa cultivar (12.39 to 15.12) during the estimation. Table 422.4. Summary of oeficient i3f detenninet n (R). RSE and linear regression weight madeIs (for Chausa and Dash ari mango cultivars) on the basis of shape attributes obtained by image processing based on color CYS. Shape attributes Regression model R-square R-square RSE (Regression (pixel) adjusted) Standard Error) C!uzasa cultivar I: y = 0.003x - 66.99 0.842 0.842 12.39 Perimeter y = 0.487x - 296.6 0.808 0.805 13.67 Max Feret's Diameter y = 1.13Ox - 244.3 0.765 0.761 15.12 Waddel Disk Diameter y = 1.2 0x - 180.8 0.821 0.811 13.18 Dasheftari cultivir Area y = 0.002x + 24.42 0.765 0.760 x1.26 Perimeter y _ O.339x - 145.4 0.772 0.767 11,10 Max 1~aret's Diameter y _ Q.573x - 33.05 0.732 0.727 12,02 \VaddeI Disk Diameter y=0.992x -97.22 0.810 0.806 10.12 172 R~- 4.535 10 C °O 104 rJ --...... 0 I, 50 2OOG0 41 Jtx1 6t tI SCJOJ La 'X 40000 AIC3 (pi b) Ar (jiixeIR) a1 250 Y=10335x• 145.34 x- osos 1 R = d77 'g 104 IN s0 . 00 ?[W} lG0D 901) 10UU 11(14 Y 54 74K, 990 11114 rt Pea~IL flii1 (ix s) PainrtQ of 1LutJc fpL b1 —1.1345ti-~ .3- 251 y-0.571x`33,055 ~ 1~G ,140 I0[' 200 254 X00 350 .14 150 :U(1 250 aIXI 5U 400 4 0 M2MFeddi2net (ji~1s} Mrile'1diauh1 {PU;Isj C' C2 rnry ? { T=1,?645X-1.880 250 (fl R==d.S?1S Q 8==0.1104 ?OC AA 150 150 100 ^ 140 NI , SQ 'jJH ?i~ 300 3 Q 00 1Vsddeldi Ei veter~p~xrl~} Waddddiskdiaoidw Cieli) d1 d2 Fig 4.2.2.3. Correlations between shale attributes jar a (a1, a2), perimeter (b1, b2), Marc Feret's (e1, c2) and Waddel Disk (d1, d2) diameters] of Chausa (1) and Dashehari (2) fruits computed with particle analysis by image processing (IF) method based on color CVS and weight of fruits measured by digital electronic balance (DEB). 173 Further, the highest degree of correlation in this wing between surface area (pixel) and fruit's weight is also confirmed by the Jahns et al. (2001) and Dorvla et al. (2012) but differ in accuracy showed by them. This might be due to the difference in produce type. They have found the highest degree of correlation (R3 = 0.9955) between area 'measured by image analysis and weight of tomato fruits. In this research they have used a standard colour camera, frame grabber and PC for image analysis to measure visible quality parameters in a- half sphere illumination of fruit in order to avoid shadows. Similarly, Dorvlo et al, (2012) obtained weight prediction equation [Weight (g = 1866 plan area (amt) — 5.9 49] of banana fruit. The plane area of fruit was evaluated by image analysis of the images captured by colour camera in a well lit covered area, This prediction equation offer better prediction for weight with an appreciable R-squared value of 6 8.89% which is less than -square value (>0.765, for surface area) obtained for the mango fruits in this study ('Table 4.2 2.5). 4.2.2.4. Relationship between fruit's shape features and volume Regression models, R square, adjusted R-square and RSE for . ralu a estimation of mango fruits of C'iausa anal Drrshehari cultuvars were presiited in Table 4.2.2.5 and the relationship between shape features (area, perimeter, Max Feret and Waddel Disk diameter) and volume have been presented in Fig 4..2.3(a1, a2); (h1, b2); (ci. C2) and (dial), respectively using the scatter plot method. From the Table 4.2.2.5, it is evident that more than E0% variation in fruit's volume, in case of Chausa cnitivar, can be explained by the models based on surface area (pixel) as independent variables whereas in case of Dashehari cultivar, variation in fruits volume can be explained about 79%. Similar, Dorvlo et al. (2012) obtained volume prediction equation [Volume (cm) = 1.8518xplan area (cm2) — 11.978] of banana fruit based on the plan area of fruit evaluated by image analysis of the images captured by colour camera in a well tit covered area. However, this prediction equation was not found better for prediction of volume because the R-squared value calculated by them was much less as R2 34.07%. In addition to above, the value of RSE in estimation of volume of Ghausu and Dashehari cultivar fruits is lowest of the area and WDD based inodels respectively. 174 tii V=OW24x+29.301 0.001&;- fi?.41A IOU R'=0.81 ? I.200 1{4 11 160 N E OH > -0i74 5000 6090 7c00 1 0L00 2ROD 4I00O 60604 SOQI]0 1( )00 Arta (pies) Arm {FLxdg) a1 04- -136.36 }0 y=0.4509X-261.56 # } R2— O.35 20p ; RI=0,7378 150 'r 15U LO 11" 104 504 ?04 S 0 I( 900 [too 500 700 900 1100 Ftt J11 affr Li- pick) Painxteroffuits(pe ds) i5 I } =1 1x-?y5-3 }'=Oi473x-2 521 R==b,~1Sl * ' l0U 1u 2OO 25R 300 380 400 .150 .00 351 300 350 400 450 1a. ret diaiue(w Lxr[3) M iFCICiW [lCi+t(pLxd ) CI Cl Y=fl-'~4S3x SU36 ?10 -' y.19M-169.% 1 250 3 p x-=4.7915 [S1 4w 1)0 lOp 50 _ - _ Sip 156 200 3U} 31 21I 22G 240 260 20 WO 320 W8ddd ikdiaurcr(pirels) J %Tnddcf,1iakdi"ct {pL la) dl Fig 4.2.2.4. Correlations between shape attributes [area (ai, a2), perimeter (bi, b2), Max Feret's (c,, C2) and Waddel Disk (d, d2) diameters] of Chauso (1) and Dashehari (2) fruits computed with particle ar~a1ysi5 by image processing (UP) method based on color CVS and volume of fruits measured by water displacement method 175 Further, more than 73% and 71% variation in the fruit's volume can also be explained by models based on fruits perimeter and Max Fert's Diameter, respectively. Based on diameters obtained by image processing of digital irna es, Miller et al. (1988) and Wardowski et al. (1997) have evaluated volume of freeze damaged oranges. Table 4.2.2.5. Summary of coefficient of determination (R2). RSE and linear regression volume (y) models (for Chausa and Dashehari mango cultivars) on the basis of shape attributes obtained by image processing based on color CVS. Shape attributes, x Regression model R-square R square RSE (Regression (pixel) - (adjusted) Standard Error) Cliiuisa cultivar Area y=O.Q03x-62.41 0.816 0.813 12.97 Perimeter y y 0.450x - 267.5 0.735 0.736 15.55 Max Feret's Diameter y— 1.061x - 2.25.E 0.718 ' 0.713 16.06 Waddel Disk Diameter y = 1.198x - 169.9 0.791 0.787 13.81 Dushe/iari cultivar Area y = 0.002x + 29.30 0.734 0.729 11.51 Perimeter y = 0.3 4x - 136.3 0.763 0.759 10.87 Max Feret's Diameter y = 0.547x - 28.52 0.721 0.716 11.79 Waddel Disk Diameter y ° 0.945x - 89.03 0.794 0.790 10.14 Furthermore, the Appendix 3 (A-3.2) displayed the tables of mean values, standard deviation (S.D.) and range (min, max) of some of more important geometric attributes (AIMD. GMD. aspect ratio, sphericity and total surface area) of mango fruits hausa and Dashehari cultivar) in real world analyzed by IF techniques using color camera computer vision system. The total surface area (mm2) dis'pIayed in the Appendix 3 (A,-3.) was calculated by adding the areas obtained from the image acquired at the 00 and 1800 orientationsIpostures. The area (pixel) obtained from the image of fruits acquired at the 00 onentationlposture, only, was considered during the regression analysis for the correlation of area ( xit) with the weight and volume. 176 4.2.2.5. GemetrIcaI attribute based modeling of fruit weight and volume Modeling of inarngo fruit's ,crass Table 4.2.2.6a represents the developed models for prediction of fruits weight using dimensions of fruits obtained by image processing and analysis of fruits images acquired by color (RGB) camera. Thirteen models were suggested for weight/mass estimation of fruits. The coefficient of correlation (R value) was recorded > 036 for ail calibratiothalidation models except mode] na. 3 based on thickness (R=0.60) of fruit. It shows that about 60 % the variation in the weight can be explained by the model based thickness. Similarly, models based on fruit length (R=0.858) and width (R=0,764) can explain the variation in weight about 85.8% and 76,4 %, respectively. In contrast to present observation for mango, Khosarn et aL (2007) predicted pomegranate mass models by applying different outer diameters and found that mass modeling of pomegranate based an min-or diameter (M = k c t k2, R2 = 0.91, RSE 15.73) is the most appropriate than the models based on major (lvi = kIa + k2, R2 = 0.57, RSE 34.57) and intermediate = k1b + k2, R' 4 0.88, RSE = 18.50) diameters. Similarly, Lorestani and Tabatabaeefar (2006) developed various models for estimating mass of kiwi fruits based on physical attributes and recommended the model to calculate kiwi fruit mass based on intermediate diameter [M _ -64.14 + 293b, R2 — 0.78). In another study, by Tabatabaeefar and Rajabipour (2005) suggested eleven models for predicting mass of apples based on geometrical attributes but they recommended the model based on minor diameter (M _ 0,08c2 — 4.74r, + 5.14, R2 = 0.89). Moreover, model no. 11 can be considered as best model on the basis of highest . R value (i.e. 0.905) and lowest SCISP (i.e. 10.81/I1.43). But all three diameters mtist be measured for model no. I1 which makes sizing mechanism more complex and expensive. Similarly, another model no. 4 can be selected as best model on the basis of lower difference between SEC and SEP and good fit (R — 0.886) of data without significant bias in calibration and validation. In addition to theses model, model no. I which is based ion single variable can also be advocated to select as one of the best model because of law value of SEC/ EP and reasonable fit (R >0.850) of data. The single variable measurement for model no 1, thus, mares easier and cost feasible. 177 Tik 42.16a. V rioi $ML1t n 15 basal a IP diTn io a1 trader[ lk eogb, bidth Ibiclmess) for ding wcight of cngo thii of L'h ri a end DaSWAri vliivat S kx1Is li SEEP ¢trclatirm No. i i it i i i i i 1 hrl ~ +~lul~l~~ culibr~lion VaILd tun Cahhdtiuu Vd1 d I m I y131U.-WAY 13{}4 133$ •0O)L) (). (] [I&S 2.K W+x M:j.JW—Ij9J. 15Th 12 I(I}1 ?I74 11751 3 199 .O1 OI6S O5 4 m= ,LSY+x~ h~=40 x10~~'-4~~5 1175 1202 i1OI}5 0189 01880 5 M-K,Lff I 1, .: 493 ai4IT—x.53 14.77 14.OR -1).t1)( 0.011 114 6 MM_ 'D+X# 1~-!~93pt ''D ~#4 I4 11.11 -0.(l j 1l.1~11 OXI 7 Nf-110d 4K1 11D-4L I2.1; 12.61 0,011 •U.017 B.S74 0,9 8 M=Ki p+X#J'i X, ME t2b —2 1a 11.16 11! ,QQ4 O1o1 .Q91 Q, vmrj~-4'f43 M=].04L+J T—L6 ,23 INNS 12,73 O. D 0.019 ' 0.875 0.665 10 MiKL+K,Lrq—K, -1 Lt2,i4YJp 1-91.16 11.17 f 1. -o R9 ll xi +KJ +ic,f+x+ m=;.t +2.3 '+4.? '-21E 14.81 11.43 OI -li?2 I9U5 0.93 12 M•Ki 'tX2LIFTtK M- ,I7L1,21:~14~I~1a~ ,903 11.41 ll.~1P 4, O .I )I M . 93 , 1 M=K( -KFL'+f N M=413'x' 'f,H`J~cl0'LFT-139. 1I,73 121 -.0J1 -001? O, 5 0,877 119 rj Similar findings have also been presented by several researchers such as Spreer and Muller (2011) have developed the correlation model between mango mass and the single dim ion such as length (L), width (Wmax) and thickness (Trnax) based on power function CM = 0.000136(L)3 0 $2, M= 0.00I ( max) 249, M= 0.002 5(Tmax)'.793s] and they have reported the coefficient of determination as 0.84, 0.92, and 0.88, respectively. The coefficient of relationship found in this study, thus, slightly differ might be due to the difference model type. In another study by Khezri et at, (2012) suggested the mass model based on intermediate (width, b) and minor diameters (thickness, c) as M ° - 152.35 + 2.38 h 1- 2,09 c with It 0.93 to predict peach mass based on outer dimensions. However, the models suggested in this study for mass measurement found superior than the quadratic model (M 5.648-0.8051,+0.036L2) suggested by Lorestani and Kazerni (2012) for mass measurement of castor seed (R1 0 771 or R =0.878). Similarly, Naderi-Boldaji et al, (2008) recommended apricot mass model based on the minor diameter 2.649c-66,412. 1 2 = 0.954) and Seyedabadi et al (2011) recommended cantaloupe, nonlinear mass models based on the fruit width (M= 2.614b239 1, R2 = 0.957, SSE = 0.118) with higher determination coefficient (R2) and the lower standard error of estimate (SEE). The higher correlation coefficient and lower SSE in these findings, however, contradicts the present study, it might be due the difference in produce type. Further, the values of SBCJSEP of developed calibration models varied between 10.8] and 18.80, and between 11.43 and 19.90 for validation models, respectively. Modeling of mango fruit's volume Thirteen models were developed and suggested for volume estimation of fruits of Chausa and Dashehari cultivar (Table 4..2,6b.). The model no. 1, based on the length of fruit, was found lest for the calculation of volume of mango fruits. Since, the model no I has highest coefficient correlation and lowest S / EP with no significant bias in calibration and vajidation. In addition, the model no. 1 is based on single variable avoids complex nature of sizing mechanism and makes it inexpensive ('Table 42,2.6b). 'In contrast, the model no. 7 & 11 can also be considered as good model for predicting volume of both mango cultivars (Chausa Dash&hare) but all three diameters must be measured for these models, which makes the sizing mechanism more complex aitd expensive which is in agreement with the findings by Khanali et al. (2007). 179 T OOO 4i1,6b mad&]l 1asif on IP dimemimoi chitirs {1 ngth, br h ii & ianp fruits of CIwt and Di. ri ciiId 'urn IAC Ib RBI JOB SEGO Bi€N CzAlm No, Cafibri',ium V~1K ii CIibct[oli i&itiaor Culihrtim V iI a iaa 1 V:L 2L-1 .11 1 .4i I6.R -O.DO 11.4I9 x,791 77 I, 7 11#57 O1 1J rO42 U,715 0.698 3 V=KIT+x7 V=3,t4T+2I.p4 1q. 8 20,16 •I.MDU 0,034 O,5 O,54 4 V .45AI IN-12.56 1$.23 18.79 O.EIO 0,07 x,722 4,1Ol 14 1934 'O.0D 4,099 UUPP 0.681 6 V: , F +xI ~~=]~1 9 -]70.60 151 IN OE 0,{ 1 0,712 4.695 ? V=K~+?,+Kd 17•{.3bfl -ts.i 1 UI .5 0.735 a l '= ,1,+ 2 ' ~=1~~t+ t L a2i 19.11 IB.97 O.N~0 0,105 0.716 0.695 9 VDK1 Lt42r•i•K, v=Zi9L+IAN-t[65k 1.~7 19.41 0.40 0,M 0.60 W4 '0 KC -2.6210 03,5,xleLw.gy. 19,21 I?,14 MO 4, X1,722 1,690 11 V—KL+K'2F+K.T+~'a V,1 ~ , 'i115T-11b~~1 I'~.3 13,5S 0000 O~~i1 .111 12 l ,l 0.X3 0.093 {1.72( 4O 13 v= ~~ + }~ + , fir_-i,97T+6.4 1 -'LJt7'~109.47 18,16 18,95 4,000 O.03 0724 , 95 AD . They have suggested 4 models for predicting volume of citrus fruits on the basis of major (14 intermediate (w) and minor (T), combination of these diameters and the value of coefficient of determination (R or square of coefficient of correlation i.e. R) reported by them were 0.90, 0.91, 0.63, and 0,96, respectively which were slightly higkter than the observed value of coefficient in this study might be due diferenoe in fruit type and fruits taken for study from two different cultivars (i.e. Chausa and Das lari). However, lowest R value reported by them for minor diameter (thiek ess) is in complete agreement to this finding which means that the variation in volume of fruits is least depended on minor diameter. 4.2.2.6. Common external defect detection of mangoes The developed algorithm for color camera based computer vision system, in this research, found impressive in identifying blemishes on external fruits surface. The spot like lesions caused by anthracnase disease and black spots arising from physical, pathological, insect or other agents that definitely contrasts with the overall colour, and which may penetrate into the flesh are found some common skin defects that causes lowering the prices of mangoes in the market. Fig 4.2.2.5 shows captured images of ripe mango fruits with slight defect (0-1 %) or with no defects and the results of thresholded defer using proposed algorithm. As demonstrated, the defects on fruits surface seen as black spots can be detected using proposed algorithm accurately, The appearance of these dark spots is found bright after segmentation and the non- blemishes area of the object appears darker (C, Fig 4.2 2.5) On hue color plane of fruits B), these (defects) are also looking brighter (Fig 4.2.2,5). Thus, after extracting hue color plane among the -various color, the fruits is fresh or defected can be inspected. But, to evaluate the severity of disease/defects or black spots which generally caused by various factors such as physical, pathological, insect or other agent&, segmentation of defects and further processing of image is needed. For instance, the Fig 42.26 and Fig 4.2.2.7 show the images of moderate and severe defected fruits. There are various scattered small black spots or lesions on the fruit's surface/ skin (A, Fig 4.2.2.8) which relevant appearance is brighter after segmentation (B, Fig 4,22,S). Since, those black spots affect the appearance of whole mango fruit because of their low pixel intensity, 181 The percent area of these black spots can be calculated by evaluating total area of holes and particle through particle analysis included in IMAQ vision software. Similarly, the severity of anthracnasa disease (black lesion) and blemishes on skin because of other factors can also be evaluated (Fig 4.2.2.6 & 4.2,2.7). Since, according to the scales of Smoot and Segall (1963) and Roomea & Jeffries (1993), disease severity can be defined mainly by either size (diameter) or number of spots, respectively. Like monochrome camera based CVS, some fruits with many small spots or some others with a single large lesion were also found which caused problems in assessing the severity of disease accurately. Fig. 4.2.2.6 shows larger size of black spots on fruit's skin infected with severe anthracnose disease. The segmentation is difficult from the background when these spots or bruise found around the edge. However, these were detected out from the images acquired from the other directions/ postures, _ Similar finding has also been reported by Chen et al. (1998) and have stated that it is impossible to detect bruise when the bruise is at the position r (bruise distance from centre) > Q,837R (radius) i.e. around the edge. In addition, Fig 4.2.2.E shows the images of shriveled fruits with black spots and appearance of blemishes on the surface after segmentation. The blemishes or black spots or lesions on the surface of shriveled or dehydrated mango fruits can also be detected using proposed algorithm. However, there is no much evidence against shriveling mango found through this algorithm except wrinkled boundaries or twist shape appearance of fruits. Perform wice of the proposed algorithm As mango fruit was rotated 90O b h reea the acquisitions during capturing of six images at 00, 900, 180 , 2700 and 3600, some defects found visible in more than one frame or posture. It was decided that the defects should not be considered again if it appeared in another frame to evaluate the performance of the system. The performance of developed algorithm was also evaluated on the basis of its accuracy and efficiency. The efficiency of proposed algorithm was noticed higher (98%) when fruits were not defected or slight defected. The inspection of 2% fruits was missed because of over brightness or non uniform illumination of fruits. 182 Fig 4.2.2.5. The images of fresh mango fruits of Chsci cultivar with (0-1%) or without defects (A), hue plane of color (B) and results of defects (), segmented using proposed method. 183 Fig 4.2.2.6: The images of fruits with severe black lesions (anthracnose) and results of proposed algorithr after defect segmentation. Fig 4,2.2,7: The images of rwngo fruits with severe black spots because of physical or pathological or insect damage and results of proposed algorithm after defect segmentation. 184 in contrast, its efficiency for the inspection of severally defected fruits was found 94% because of overlapping of defects and background. In addition, the efficiency of algorithm for shriveled fruits with black spots was found lowest (83+3%) might be due to lower avenge pixel intensity (64.89 to 98.52) of shriveled fruits. However, the average pixel intensity of fresh or very slight defected fruits was found between 105.23 and 129.51 at same acquisition conditions. Table 4.2..7a: Calculated efficiency of various defect types Defect type No of fruits System Inspection — Efficiency True False Fresh or Slight defected fruit sU 49 1 98% Minor defected faits 50 48 2 96% Severe defected fruits 50 46 4 94% Shriveled fruits with defects 30 25 5 83.3% Total no of fruits 180 168 I 2 93.3 % The overall efficiency of algorithm iWas found 93.3%. Thus, using the proposed algorithm 93.3% images of mango fruits of Chausa and Dasheliari cultivar can be processed and segmented for defect detection (Table 4.22.7a). Similarly, the accuracy of proposed algorithm was also decreased as severity of blemishes or disease infected uea on fruits surface increased (Table 4.2, 7b). As fruit changes from no defect or slight defect to severe and shriveled, the accuracy of system decreased from 96% to 87.5% and 73.3%, respectively. In spite of this, the overall accuracy of algorithm was recorded 88.6°% Further, the average inspection time taken by algorithm for Dcishehari cultivar were also evaluated as 19.75±0.84 ms, 23.47±0.12 ins and 25.04±0.13 ms for fresh or slight defected, minor defected and severe defected fruits, respectively. The average inspection time for Chausa eullivar was noticed slightly longer than Dus ekari cultivar and it was found as 29.28±0.11 ms, 30.280.08 ms and 33±0.15 ms for slight defected, medium defected and severely defect fruits, respectively 185 A B A B Fig 4.2.2.5: (A) Various images of mango fruits with different common min-or defects (6-9%) on skin in different posture, and (B) results of proposed algorithm after defects segmentation. The bright spots (B) on surface are related defects. 186 Fig 4.2.2.9: The images of shriveled mango fruits with black lesions and results of defects segmented using proposed tiithod- 1S7 The longer time taken by the image of Chusa cuttivar fruits might be due to the variation in fruit size and peel color. The weight of CJausa fruits used in this current research was ranged from 130 g to 250 g and the weight of Dashekari cultyvas was between 130 g and 212 g. The pixel intensity of fruits of both cultivar also differs slightly and it was noticed that maximum time dtuing inspection wns taken by step: color plane Ihue) extraction. The peel color of fruits, thus, affects the processingfinspection time of image by any algorithm. In addition to this, larger the fruit size, larger the number of pixels and thus larger the processing or inspection time taken by the image. Hence, the performance of a gorithm can be evaluated on the basis of its of iciency of inspection, accuracy of true defect identification and total inspect time taken by the image. Table 4.2.2.76: Calculated accuracy of various defect types Defect type No of fruits System Identification Accuracy True False Fresh or Slight defected fruit 50 48 2 96% Minor defected fruits 50 47 3 93.3% Severe detected fruits 50 44 6 87.5% Shriveled fruits with defects 30 22 8 713% Total no of fruits 180 161 19 85.6% In addition,'as the storage period increases the sire (diameter) and numbers of spots on surface also increases. The image analysis technique can also be used to monitor accurately the effects of ripening and storage conditions. Consequently, the fruit sorting based on this novice method certainly benefits both the producer and consumer, 4.2.3. Non-destructive quality evaluations of mango fruits using UJ computer vision technology 4.2.3.1. Size measurements Actual !eng h, K'idth and Ihkh ess of mango fruits Mango fruit's the physical properties (Iength, width and thickness) were measured manually by means of caliper and by image processing method lased on UV computer vision technology presented in the Table 4..3.1a along with their summary of descriptive statistics. Mean ± STD and range of length, width and thickness of Chausa fruits measured by image processing were 96.23 t 7.38 E [78.83, 110.98], 57.06 ± 3.39 rs [50.83, 63.81], and 48.97 ± 4.23 E [38.37, 55,99], respectiveiy and ofDashehari fruits these were 94,2917.79 t= [77.51, 106.80], 55.51 *228 E [49.50,. 60.19], and 46.11 2.21 e 142.11, 51.211, resp livel . However, mean ± TD and range of Chausa fruits length, width and thickness of Chausa fruits measured by caliper were found to be 100.02 ± 6.70 E [85.80, 113.92], 58.90 ± 3.92 E (50.52, 64.98), and 50.64 f 4.59 E [33.59, 58.69], respectively and of Dashehari fruits these were 102.63±7.78 E [85.45, 120.0I J, 57.66 ± 2,55 t= [51.87,6395 and 5283±3.14 E [47.27, 60.13], respectively (Table 4.2.3.1 a). l rom the Table 4.2.3.1 a, it is evident that the mean and range of length, width and thickness of fruits of both cultivars measured by image processing technique ('When fruits, image acquired by UV camera in IJV light) were found considerably smaller than the values obtained by digital vernier caliper. In contrast, the differentiation of the descriptive statistics such as standard error, standard deviation and sample variation between image processing and caliper measurement is unpredictable. For better observation and understanding, Table 41.3 , lb presented the resu[t . of the comparison of measured (VC) and computed (IP) these physical properties of mango fruits in terms of the minimum, maximtun, and average, values along with their scale factors (k) for both cultivars. In addition, Appendix 2c (A-2c. 1 & 2c.2) shows the complete description of the proposed algorithms for physical characterization of Chausa and Dashehw' fruits on Lab VIF.W software used in the ultra-violet computer vision system. 189 Thk 412311,3uuiiiy slatistks of lbe m red val of go wit's Leiig Bead' artd'hikne 'i UY cams is it CV was oscd. .r r. ,- ~. -~ -• T r r~ 11- Dmrd?tlY( mtlifrs RpIr.a1 prop~rtiis mp.murt 1 byim p. prnr~asing Chifrao 1 It Dahrkri fruit ~angC (mm) Brc9 lh (mm) Thi4kncss (mm) Liigi1 (m) Brcadlh (mm) llriclmes ( m) 6.23 X7.46 4:7 55.51 46.11 5!d E 1.03 0.4, 1L52 1101 SL it 4.3 739 2 2 2127 Simile w 54A3 10.94 14.01 57.54 8.32 5.16 Min 7&83 U :37 71,51 '1MRR 12,I1 1 1O9 3I91 S T A I). 19 SL2I Ph io! pr rhos ieiiiircd by { aiiper (maiir) ~.--- [iiI11 it Dimension Dwriptve h~ bia St]drd error 5tn~ddd D a1i ft S iiip l V ria on Mithmum M nu n Lol(b., vin 100.02 '394 6.~0 44,9] $ .$0 113.92 Badth, m m 58.1]' 0.55 3.92. 16.34 5',52 64. Th ess, ima ~ 0.9 4I5~ 21 I':, tl e!iiri tol~1vaT A l: n~lh,rnm I(12,63 1.03 '1.'11 61.19 120(1] Br tl m 57 6; 0.47 255 11 ' 5118 Thiclrnc~s, r m . ' 0,42 3,14 .9. 4 47.2 0.13 lip ~~ti 11f 1 1% Tah1E &2L .li~.,C cif' pir1(VC) and c1impod iP) ]crgth, n'.idth tike of mollgo hib ioo in t)i 1~rrtl~ I, Sa1~ fir th 1'llm~s 1~ 'ultivar~ VC IF f1dor C) VC IF r (k) 'C I boor An 5.52 X0.85 290 33.58 $.3' 2.9b l ax 11391 119 19 6321 8. 559 AVE. 1.t2 ? RA 57,% (1,64 497 I ~ k1Lu s 5-d( 17,51 2. l 51,K? 49,51 47.2a 4211 2I9 1x l ~.'1 6 691 0.19 x.19) 51.21 Avg 1O26 ?,i 55 IS1 52.12 46.11 Tab1c 42i2 PuhTd t-tcst f1tly5ia; for f uii's kn ih. br~w1t[i and ±ucic of Charsa uwnt cu1d r, I tw t n Ium meunmcnt i ietho h [vemid caBper amd coluput (EP)] DIr WA iie R Stand~rd 16o SW ror of 9% Cl Tr iIi 9% of if•t t !ev2I1mm diffrw diffirlpec nIR~o diU frlpe4` A m'u (mrn) (mm) (!lu ) Cfiiai. i ul(ivar Lmo 0.00 Th 2.A6 3.045 4.l i 3,'J9'42 1Ldth (1.O1 I9O4 L. [. 3 I (I ?366 1.36 2.32 -1.5 5,7Q 1 ickn ss (M,O 11S i 7I)(I4 17O9 4,2184 392,35 •L5; 520 1b l~arr! calr 1 ngth .954 2416 8.iS17 42:8.7( 3r~adth O.m U. J1 L93 4011 01587 ;6 17; 6,40 Th kr 11 ').913 tI766 .2i7 U.1R 1 2.U5:2,68 3.80:8.M --1 r.1 r r rrttr- -r-r-- 141 Coinpur£sarn between image prncessirrg UP) and copper (manual) melJwl The predict d outer diameters (major diameters: length; intermediate diameters: width & minor diameters: thickness) by image processing and analysis of images acquired by IJV camera in UV 3igbt were compared with the actual (DVC) diameters measured by caliper. The comparison results between lengths (a, d), widths (b, e) and thickr esses (c, f) of Chaa and Dashe/rari cultivars, respectively, obtained by both methods is shown in Fig 4.2.3.1 by scatter plot. Higher coefficients of determination (R), about 0.96, showed that the both methods differ slightly (4 ') with respect to length of fruits of both cultivars. But, this difference was higher for width and thickness of fruits between the methods might be due to the fact that the diameter of the corners of the fruit is not perfectly round and hence measurements depend on the point of view, This is in agreement with the findings by Van Eck et al. (1998). In spite of this, the observations of fruits width and thickness were consistent and the coefficients of determination of these diameters between both methods were found reasonably high. The R2 values of width (0.904) and thickness (0.906) of Chusa cuttivaz were found almost same. In contrast, the R2 values of width (0.901) and thickness (0.913) of Dashehari cultivar fruits were found narrowly different. Further, in another comparison test conducted between predicted (by image processing, 1P) and actual (by digital vernier caliper, DVC) outer dimensions (length, width and thickness) by means of the paired 1-test, both methods were found statistically significant (P<0,05). The obtained P-values were recorded less than 0.01 for both cultivars. Table 4.2.3,2 displays the mean differences between length (L), width (W) and thickness (I) measured by caliper and by rage processing methods, For Dashehari ctiltivar, the average difference between outer rnensions (L, W, T) by both methods was found higher i.e. 8.06, 4.04 and 6.28 than Chausa tivar (3.67, 1.83 and 1,87), respectively, Similarly, the corresponding standard deviations )) for length, width and thickness differences of Chausa fruits were 2.09, L68 and 1.70 :le for Dasheha;•i fruits, the corresponding standard deviations for length, width and laaess differences (SD) were 2.40, 1.19, 1.27, respectively. The values of standard 'ation measurements, Table 4.2.3.2, revealed that the objects were varying more with cot to length than the width and thickness of fruits of both cultivars. 192 cif 1 y -5. S 1 110 R°= O.S4 , 0.9 * pW 4 loo • 90 94 100 110 120 130 Sly 90 100 110 120 hLhtISQ kith. rni{VC a 65 - v 0. r 63x 112.(8 b '~1.004x-4-107 pa l - =0.904 1=0,901 ~ 4 as 45 4S 4O I 4 5$ 60 65 70 S 60 65 Mw width.Mb (VC) I 11' )dd0l,]IIRl(C) b e - r=0,816 +1. 55. }0 1x+1116 R==Q-913 R-'= O 4 S. o ~ 40- u ~ 4 40 40 45 a0 SS 60 65 h ft fi k}1e , umt (VC) Mi i 20 ihic :nes3.nnn ('C) C f Fig 4.2.3.1. Comparison of computed and measured outer dimensions (L, W, T) of Chausa (a, b, c) and Dashehari (d, e, f) fruits with image processing (I) method based are UV CIS and digital vernier caliper (DVC). 193 L8 OF Camera Id UV camera -ISDf1G 14 ° $ +1 96 SU , 12.8 O pP O a pC P 4 $°° a n a4 a 7~tC-0o ~P a C ° a ] 7 x o 0 qa¢ P p° 4 ,.,`. B,L 4 a GO a g a A O ~ 4 -1SD-96 4 4 a . L'.6 5> -0-4 ° 3-4 M 0 4 95 1D IOd 110 115 12 84 90 IGt'I 110 120 Avtrn@61en lh (VCM - LP) ArarMV Un j1h {YC(b - I?) a Y TJ t$meru liv camera a o a 6-4 a 8 +I 9d 5D $ # 5-s I • 0- 4 4 O O MCgn a h Mean C1 . 4 p8 J d D # yy ¢ 06G p 2S G O Q 4 d a ° 4 Q 0 ° 4 8 -I. FS 1Q 0 0 Co 0 ¢ -2 4 •j. LI 2 O O 1.7 50 53 60 b5 70 46 50 52 54 55 •39 60 C2 54 Averwg dtb (VCM - 1F') - 1r1 b e UV camera ° ¢4 1a UV Cerra 41.46SD +1.96 SD 00 5 a a 6 8.6 4 • C o d p P 4 O 8 ¢ q a ° 7 4 • o # a ° Memo I4dE671 8 1.9 a b ¢ 4 ti $ 6 9 a P 0 ° ° " 4 . ° 4 ° -I.9C* SD L o 1.96 5D . I_ 42 47 52 57 B2 as 45 SD ss n Avre d ki [ r4 i -' IF1 ATerage Iht#][otas (o I • 1F`] C - Fig 4.2.3,2. Bland—Altman plot for the comparison of outer dimensions cL, W. T) of Chawa (a, b, c) and Dathehuri (d, e, f) fruit's computed with image processing (IP) method based on UY CVS and measured with vernier caliper method (C.M); outer lies indicate the 95 % limits of agreement and ceziter line shows the average difference, 194 Furthermore, to reveal a relationship between the differences and the averages these measurements, to look for any systematic biases and to identify possible outliers, the Bland- Altman plat (Bland & Altrnazx, 1986, 1999, 2007), or difference plot was also used. Fig. 4.2.3.2 shows Bland-Altrnan plots for length (a, d), width (b, a), and thickness (c, of the faits (Chaos!, Dash iiarn), respectively. In these figures, outlier indicates the 95 % Iin it of agreement and the centre tine shows the average differences. The 95 ,% limits of agrreement, for comparison of length (a, d) width (b, a) and thickness (c 1) of the fruits (Chausa, Dasheho-r), computed by image processing (LP) method when fruits images were acquired by UV camera in tJV light were (-0.40; 7.80), (-1.50; 5.20) and (-1.5; 5.20), and (3.40; 12.80), (1.70; 6.40) and (3.80; 8.80) respectively (Fig. 4.2.3.2 and Table 4.2.3.2). From these results, it can be stated that mango fits outer dimensions have no effect on the accuracy of estimated length (P>0.05). 4.23.2. Estimation of shape attributes by image artdlysi Four size d pendent shtpe attributes of mango fruits were evaluated to differentiate the eultivars. Table 4.2.3.3 displayed the values of fruits elongation, compactness and circularity (Heywood) and type factors for both cullivars. The Dash€hart fruits were found to be more elongated and coact than the Chzusz fruits. In addition, type factor of Dashehr ri fruits observed higher than the Chan.ga fruits, Thus, the elongation, compactness and type factor can be 6sed to differentiate both eultivars. In contrast, the mean value of Heywood circularity factor of fruits of both cultivars evaluated by image processing in case of UV camera image acquisition setup was found same and cannot be used for differentiation of fruits type. The mean value of elongation, compactitess, circularity and type factors of Chausa fruit were 2.23 ± 0.32 E [1.64, 2.74] (CV, 14.4 °o), 0.75 + 0.05 E [0.63, 0,80. ] (Y, 6.4 %), 1.09 ± 0.06 E [1.02. 1.42] (CV, 5.7 %) and 0,96 + 0,05 E [0.84, 0,99 J (CV, 4.7 %), and of Dashehari fruit were 236 ± 0.40 E [ 1.72, 2.95] (CV. 17.3%) 0.78 0.{t2 (0.70, b.S ] (CV. 2.4%), 1.A9 ± 0.05 S [1.02, 1.17 ] (CV, 4.1 %) and 0.99 f 0.01 E [4.91, 1.00 1 (CV, 1.3 o), respectively. From the Table 4.2.3.3, it is evident that the fruit's elongation is highly variable factor than the circuIarity and compactness of fruits. Further, the other shape attributes such as fruits surface area, perimeter. Max b'eret's and Wadclel Disc diameters were also evaluated and correlated with the mass and volume of fruits. 195 Table 4.2.3.3. The mean values, standard deviation (S.D.) and coefficient of shape geometric attribttes of mango fruits (hausa and Dashe!ari cultivar) analysed by 1IP technique based on Uv cvS_ Cu1dvaa Parametcrs Minimum Maximum Mean S.D. (±) C.V. ( ) Citusa cultivar Elongation Factor 1.64 2.71 2.23 0.32 14.4 Cen'pactness Factat 0.63 0.80 0.75 00S 6.4 Heywood Circularity Factor 1.02 1.42 1.09 0.06 5.7 Type Factor 0.84 0.99 0.96 0.05 4.7 DasheIiari cultivar Elongation Factor 1.72 2.95 2.36 0.40 17.3 Compac.tnes Factor 0.70 0.83 0.7$ 0.02 2.4 Heywood Circularity Factor 1.02 1.17 1.09 0.05 4.1 Type Factor 0.91 1.00 099 0,01 1.3 4.2.33. I elatiotsship between fruit's shape features and weight Once shape fb2h yes (area, perimeter, Max Feret's and Waddel Disk diameters) have been extracted from the fruit profile images acquired by U camera in UV light by image processing method and they have been combined linearly to fbnn an estimate of weight. This linear relationship is a model for the relationship between shape features and weight. From the Table 4.2.3.4, it is evident that these models are reasonably fit to the data. The surface area (pixel) and Waddel Disk diameter of the Emits of both cnitivars show about 86.0 % variation in the weight is explained by the models. In addition, the RSE in estimation of weight using these models were found lowest. The finear methods provide very fast and simple weight estimation lased on some simple reasoning about theapproximate relationships between the shape features. Using the scatter plot, the relationships between the shape features and fruits weight are shown in Figs. 4.2.3.3. 196 '50 v= 0.04%-68.05 350 - y=p.F7{I&x-S-M • 150 J I 10 fF 104- 50 56 > 0 .aoao r5o0 3aaio 350!0 40O N 10H0 15040 3101X ;50'10 4 M1O0 Arta {pixels Arcs {pig ls) al 350 Y = Uj? -1F, 5 '.5O - }•°0„i6hi-65.93 R2=0.7667 200 150 1 100 J 50 SQ :~ 4 600 '00 Soo 50~l 610 X00 8OO P,' nItteroffnLitF(Fise1) Paimetr ONTdIs{pixelr} h, '-0 1 r., :.5o }' = 4, b.13x+ 6.181 y=a,39R-61.39 100 100 154 ?00 250 300 34 ]'U 17n r 1 ?;D .U(] 3;4 Ma Ft d dianxldr 1pi~tlx1 11a?~ Ferd kmieta (j Lx {x) cl C2 — .. ---.. u ?50 ;.17fie. 9t.O 250 , 1 S66x-187.3 R0.X4 7 , ~ R7= 0.559 E 1 5 SCE , 100 I50 200 ?30 100 150 00 ~50 Wadd 1 di_*C diemdq{pL'cl:~] 1'iriel did li~ iiier (17i e1R) Fig 4.2.3.3. C orri 1ations betweeF shape attributes [area (ai, a2), perimeter (b1, b2), Max Ferct's (ci, c1) and Waddel Disk (di, d2) diameters] of Chausa (1) and Dashehari (2) fruits computed with particle analysis by image processing (IF) method based on UV CVS and weight of fruits measured by digital electronic balance (DEB). 197 Table 4.2.3.4. Coefficient of determination (R) and linear regression weight models on the basis of area (pixels) obtained by image processing technique based on LTA CVS. Shape attributes Regression model R-square R square RSE (Regression (pixel) (adjusted) Standard Error) Cha&sa cultivitr Area y = 0.007x - 68.05 0.864 0.862 11.48 Perimeter y = &532x - 184.5 0.760 0.756 15.28 Max Feret's Diameter y 0.895x - 6L9 o;BQo 0.796 13.9$ Waddel Disk Diameter y = 2.378x - 297.0 0.848 0.846 12.15 Dashehari cullivar Area YJ0.00 x-5.998 0.854 0.3 52 8.88 Perimeter y 0.3 1 x - 65.93 0.714 0.709 12.41 Max Feret's Diameter Y 0.633x+6,181 0.744 0.739 11.75 Waddel Disk Diameter ys1.8S6x-187.3 0.859 0.856 8.72 4.2.3.4. Relationship between fruit's shape features and volume The fruits of Chausa and Das#hehari cultivar were photo raphed using UV camera image acquisition set up to acquire the digital profile images and the surface area, perimeter, Max Feret's and Waddel. Disk diameter were measured by image processing to estimate the volume of mangoes. Highest coefficient of determination (R-square) was evaluated for surface area of fruits followed by Waddel disk diameter (R-square). As on the basis of surface area more than 3% variation in volume of Chuusa fruits and 8 L% variation in volume of Der. he sari fruits can be explained and the bigger size fruit is considered of better quality. The area based model can be treated as good model for the estimation of fruits volume. Similarly, on the basis of Waddel Disk diameter more than 81% variafizn in volume of fruits of both cultivars can be predicted with Lowest standard error in estimate. The Table 4.2.3.5 displayed developed linear regression models, R-square, adjusted R-square and standard error of the estimate. Models for volume estimation of DmshehQri eu.tivar 'were having lowest standard error of the estimate. The scatter plots between shape attributes and fruit volume are presented in the Figs. 4.2.3.5. 198 250 - Y=0A77x 3,4f ?5f? y4]D{c+x.9= R': 4,35 E 211U 4 ;:iet4# 1Q~ 1,0 J4 20000 25404 30000 A5 0 40000 4 4O U4 Iowa,Fe a pp"Q61 3000 :lln (pi ts} ai a2 - ~ 130 1.4 .' Q~ = 106' 4 100 ?OU 600 00 %0 500 600 700 804 Pe ter offlair{picds) Flvnetf'aiLvi[s fcli} b2 0 y=4-605"C 18.727 R=0.71 1 iOc SO SO > 100 150 200 !0 300 150 L!11 170 2,U 2-0 320 3 k I id di ndnr [pads} 1~c[Fcrtdiildc {pc[cl~ 2i4 , y1.74•1105 2c j 1\•-0-817 15 100 50 100 50 104 ISO. '_0Q : Q too 150 200 '50 Waddddi-kdi3n (putt) Va ldel disk diamder (pie~sl dl da Fig 4.2.3-4. Corrc1aioiis bemeen shape attributes [area (a', az), perimeter (b1, b2), Max Feret's (cl, c2) wid Waddel Disk (d j, d2) diameters] of Chq Asa (1) and Da ehari (2) fruits computed with particle analysis by image processing (M method based UV and volrame of fruits measured by water displa meat method 199 Table 4.2.3.5. Coefficient of determination (R') and linear regression volume models on the basis of area (pixels) obtained by image processing technique based on UV CVS. Shape attributes Regreion model R-square R-square RSE (Regression (pixel) (adjusted) Standard Error) Chausa cullivar . Area y — OM07x - 63.04 0.835 0.832 12.28 Perimeter y = 0.494x - 165.8 0.700 0.695 16.55 Max Feret's Diameter y = Q.8 0x - 48,23 0,715 0,710 16.15 Waddel Disk Diameter y = 2.261x - 220.4 0.817 0.814 12.93 Dashehari cvltivar Area = 0.005x + 0.592 0.812 0.809 9.67 Perimeter p = 0,342x - 58,48 0,695 0.690 12,32 Max Feret's Diameter y = 0605x + 8.727 0.735 0.730 11.49 Waddel Disk Diameter y = 1,774x - 170.5 0.821 0.818 9.44 In addition to above, some of more geometric attiib, tes such as AMD GMD. and aspect ratio, sphericity and total surface area of mango fruits (Chausa and Da he/iari cultivar) play an important role during machine design for the sorting/grading were aIso analyzed by IF techniques based on ultra-vio'et computer vision system. Their mean values, standard deviation (S.D.) and range (min, max) in real world are displayed in Appendix 3 (A-3.3). For the total surface area (mm2 ), the, area obtained from the image acquired at the 00 and 180° orientatiousfp stiires were only considered. Similarly, the single side area of fruit, obtained from the image of fruit acquired at the 0e oriel tation/posture, was considered for the correlation [between area (pixel) and weight (g), and area (pixel) and volume (ml)] study in this research, However, the areas in other posture (00, 900, 1800, 2700, 360°, top and bottom) of fruits were also evaluated but not entertained and reported in this research because of overlapping of one area in more the one postures. 200 Table 42.36 Vñiis MLR l on IP dirnnsi sl ha ri .s ~l bah aid ihicknss) fo detrrn[ni &gl t of ni go friit 1Catisa mnd l J ri cii]tiv r~ od1 > 1 t«u — — 9ECLSFJ (dIkii Ni. 4iI Irikin + iatlad Cl]I n is i V i t aUon C 1i i .n V iritl~ -~.~t•rr1 1 i•r r•ir r~•.1~ 1 m+K, _ ~— ~.~ 13i7 t3 •1i OR Fô I6,;9 16.;6 -0.000 D,043 0.77 0.765 3 M:X T.'. C. MONT-12.35 14.3 19.4i aIf]~Ju 0.017 0.02 016}3 4 NJ=, ~L '.L , i=3541 ~~ —]1.00 11.70 I1.~7 MOO 0.022 . 7 O, 82 M —KLT.K, ] —3,%lola-m0 1112 1152 .O,[ O 0,00 I 76 0•8]LI 6 1 •L i+ 1M=5.13XIV LWiM 14111 1 11 1219 O N 0025 1j79 5173 123 L400 r 301 UI 9 O S14 i~1~+.a t M■.I~~~~~.~s 97 9.$$ I L IIJ13 924 4,91] 9 M—X1 L- + . 3~G4.UW' IR } 111 11.56 0,000 10,014 *0 [U M:X1 L-K3 T+ 11.14 -0.400 HIS x922 0.917 11 M -K1LiJ '+K '+Z4 M■2.4IE I I. U1,2.2 ). -OMth -0.027' 0.32 0.24 2 97 10 ] 1 0,044 Q,5 4,424 U17 3 !4i= yG+l f 'i'+k, .51L+11 'Lr7-51 36 49? UUU ,929 0.923 14 M+X1 L K E1 +Kl M6,014L+48$ d -220.1 2 7 09,64 O, NN .0,09 I1 0.425 I { ( K !+K,r~ +I; M ■I.5 L+3.91D —Il&1] ,l6 491;3 •IU)IJIJ - 201 4.2.3.5. Geom trJcal attribute based modeling of fruit weight and volume Model'ing ofinangofrizil's mass The acquired images by UV camera were proecsed, analyzed and thus obtained fruit dimensions were used iii modeling in order to predict the mass and volume of fruits. Fifteen linear regression mass and volume models were suggested using one, two aid tree variables and some derived variables such as aratbmetic (Da) and gcometrie (Dg) mean diameters, aspect ratio (Rn), and sphericity (p) of fruits were also tested. The models had higher cueilcieat c;orrelatic n displayed in the Table 4.2.3.6a. Among these, m els (no. 11, 13, 14 and 15) based on three variables had higher coefficient of correlation with lower values of SEC and SEP with no significant biases between calibration and validation. But, the main drawbacks of these models is dependency on all three. fruits diameters (length, width and thicl ss). in other words, the measurements of all three dimensions of fruits not only make sizing mechanism complicated but also would be costly (Tahatabaeefar and Rajbipour, 2005), Tints, the model no. 1 based on single variable i.e. length of fruits can be advocated to select best model for the estimation of weightlmass of fruits. In addition, model no. 10. has correlation coefficient (R value) nearest to best model Qn the basis of highest R-value, can also be tested and treated as one of best model. Modeling ofofmango frvil's volume Mango fruits volume was measured using water displacement method and correlated with the outer diameters obtained by image processing methods when fruits images were acquired by LTV camera. These models have been obtained for mixed varieties. Volume model with major diameter (L) of mango traits had strong relationship among the single variable models as well as amoag the multiple variable models with higher coefficient of correlation with minimum SEC and SEP without significant biases in calibration and validation (Table 4.2.3.6b.). Therefore, the mango fruits major diameter is a good parameter for sizing. In addition, model no. 11, 13, 14 and 15 can also be treated as good models on the basis of reasonable coefficient of correlation i.e. 0.777, 0.772, 0.772. 0.775 and 0.775, respectively. The strong relationship between these variables and volume of fruits, however, were not obtaiied, it nugbt be due to the nieasuremeuts obtained from the two mixed ~rari tics. 207 l'alrk 4113,6}1. V uXhf1R .t 1r; I24e1 i IP ciroo lorial (lt th, bru1b:ind thickii) tr tr~rmini volilroe of Planar 'nits of 'hagso Dad s r~ l~ims Sr !I F1 i CUrFthit1an _ N. Calib tidn V lik in C&aoa Vali tin d:br~non V4i+1+iu I V=P-tA: V: 6 L-99.24 16.55 I?.111 (1OLI 0.414 Q.7, 0.770 2 V=.'1 iV t v _ &I6ww-136Isu l4 18,76 -O0 ] ,U ? 9I0 .714 '-. -,347 [7,48 17.71 L4 4,049 0,705 4 V: ~ Ll +~;, v =l ax t -~ i~r+15.66 18,6? I9.1 0I{l J~ •0A01 017 QI 7 5 4 - K,U+K3 =32xp~t X13 ~ 1741 !7181 0.1100 01 tI75 (i7$ 6 '=C1 LWT1K, _ . I ~4 t r+ t~ 1M.ME 1?133 (1i7 R 4.75 1 =1 J +K 'r=$,fl 9 -19510 1617 1721 4.15 0.771 475 16,11 l7.3) 11i 4770 J754 E+ =91 L-K;l-K, V=t.54t+ .l-14.6S 18 5 :9.37 OI~1I0 .0.(06 (,7117 U. '19 10 V-41 L+N,T+K3 v=1161+a T-I5a.6[ 171U ]I OR 0,042 0,759 Q.75 1 ] 6.&J 17156 0.9 r4,0 1.J.'U 0.746 12 1t+i~r+Uri+94 11= 67+107 [4fIj—SH66 16LK 17165 1011 -01771 0,742 + I Y=1C ~t +[~ v.1lt} ,lr l~ 4 r. - sa 16.75 1J3 U19 01771 0747 ]4 V=K,t+KID4 +1, V=464''6127,.-19524 15 17142 1WX r01 01775 x,751 :F5 V 1 +Ktrrr F91 v-1, ltt l~lr~-2ISI.I 16.0 1141 0.000 •024 0x775 0.751 x1.1 r.- •rrrl rr tt-•--- p3 Furthermore, the value of coefficient of correlation and value of standard error of calibration and prediction volume models based on mean diameters (arithmetic and geometric) are displayed in the Table 4.2.3.6b. However, the model based on aspect ratio and sphericity were not entertained and also not reported in the Table because of lower values of coefficient of eorrekxioii alongwith higher SEC and SEP of calibration and validation. Conclusively, it can ba said that models developed for estimation of fruits volume based on measarments obtained using UV mera showed only reasonable fit but consistent correlation. - 4.2.3.6. Detection of mango fruits defects using tV computer vision technique Ultraviolet light interacts with agricultural produces in a unique way, enabling features and characteristics to be observed that are d1 cult to detect by other methods. Ultraviolet light to ids to be strongly absorbed by many mate ials, making it possible to visualize the surface topology of an object without the light penetrating into the interior. Because of its short wavelength, it tends also to be scattered by surface features that are not apparent at longer wavelengths. Thus, ever smaller features can be resolved or detected via IJY light scattered off of them. In this work, five band pass filters of 200 nrn, 250 nm 300 nn 350 nm and 400 nm wavelength were tested for defect or bruise detection on the surface of mangy fruits by capturing their images. The band pass filter of 400 aim wavelength was found more suitable to detect surface defects or bruises. The Fig 4.2.3.5 shows bruised areas because of pressure during handling (Ai, Bi and C1) in color image and the appearance of these injuries or mechanical damage in the image (A2, B2 and ) acquired by UV camera in UV light. These mechanically damaged areas are seeing darker, rough and fall in surface topology (A2. B and :). Sometimes, it looks like burrs paper (B2) might be due to the crack on flesh/tissues beneath the peel of wit because of shock during handling. 204 2 Fig. 4.2.3.5. Appearance of mechanically damaged areas on the surface of mango fruit in the image captured by UV and color cameras 205 In the same Fig 4.2.3.5 the black spot ICI on fruit surface also found darker (Di) in UV image. The darkness appearance on fruit surface is may be due to absorption of light. More the abaauptiun of light results larger dark area.. For instance, the fruits since with anthuacnuse (black lesion) (image At and B1) and stem-end rot (image Ci and Dt) in Fig 4.2.3.6 can be noticed as darker, fall in topology and with punctured tissues (A2, B2 C2 and D2) images in Fig 4.2.3.6, is an agreement to the LTV light principle reported by Richards (206). He has reported that UV light tends to be absorbed strongly by many materials and makes possible to visualize the surface topology of an obj-ect without the light penehating into the interior parts. However, surface features that are not apparent at Ionger wavelengths, can be scattered by UY light because of short wavelength. Sometimes, the upper surface of when anthracnose (black lesion) is slightly dried it appears less dark, . fall in topology and wider cracks on surface than in wet conditiQrt might be due to slight reflection of %JV light from dry tissues. Fig, 4.2.33 also represents the color (ROB) image of fruit having anthracnose (black lesion) and its relevant appearance in the UV linage. In this image that acquired by IJV camera in UV Tight, it can be noticed that fruit surface have discontinuities and rough surface. Unlike the area with dry anthracnose, the appearance of wet anthracnose area is darker. In contrast, the appearance of fungal decayed area on the fruit surface, in same Fig 4.2.3.7, is - seen brighter. This may be due to the strong reflection of UV light from this area. The refiection of I.JV light from some other than fungal decayed area was also observed in images of shriveled fruits that results- bright appearances of these relevant areas (Fig. 4.2.18) than the fresh surface areas aught be. due to dried tissues or cells or flash, The above findings are close conffrin ty to the resz!lts roosted by Slaughter et al. (2008) for citrus fruit. According to them visual inspection of citrus fruit under UV illumination could be used to remove fruits with fiingal infection. They found that the freeze- damaged citrus peal tissue associated with fugal growth has a unique and high contrast color compared to undamaged peel tisstue, However, the fresh or unburies mango images, captured by UV camera, were also seen light dark when illuminated with 1JV light. This may be due to the absorbance without diffusion of UV light by biological tissues. Slaughter et at (200) have investigated the long wave IJY 1i ht fluorescence to detect the freeze damaged oranges using. this phenomenon. 206 LA Fig. 4.2.3.6. Appearance of severely defected fruits with anthracnose (Black lesion) and stem-end rot in color and UV image. 207 i Fig. 4.2.3.7. Appearance of Anthracnose (Black lesion) and fungally decayed mango fruits in IJV and color image 208 Fig. 4.2.3.8. Appearance of various shriveled mango fruits in UV, color and cut away image 209 Furthermore, in this study it has been found that UV camera was sufficient for collecting image data at 400 nm wavelength band required for the detection and evaluation of the latex stains on fruit surface. Various latex stains (At, B1 and Ci) on fruits surfaces are detected using UV imaging system and are shown in Fig 4,2.3.9 (A2, B2 and Q. The appearance of latex stains on fruit surface was dark (Az, B2 and C2 in Fig 4.23.9), These f dings are found to be in accordance with Nagle et al. (2012). They have reported that the latex stains can be easily detected and evaluated under -light using a low-cost web cam while the authracnose infectimn can be detected at earlier stage of disease development by measuring under UV- light. Thus, the efficiency and accuracy of the sorting and grading by hand can be improved by integrating some low-cost sensing technologies. For instance, the use of U -A illumination can facilitate the early detection of anthracnose in hand sorting, Another potential advantage of the image processing technique based on LTV computer vision system, it could be implemented fairly and quickly by human inspectors in exiting black light inspection rooms in packing lines, requiting only a small arnoult of additional training and no akIitional capital expenditures. Currently, visual inspection is done in black light rooms to remove fruit with fungal infection where training inspectors can identify fruits with small yellow fluorescent spats that should be feasible. So, this technique has been found suitable and proficient for both visual inspections using IJV illumination and black light. In addition, it is also useful for automation using machine vision techniques by the researchers. Various defects and their appearance in LW and color images have been summarized and presented in Table 4.2.3.7. 210 I Fig. 4.2.3.9. Appearance of latex stains on fruits surface in color and T T images 211 Table 4.2.3.7. Various defects and their appearance in T.JV and color images S. Bruise type/objects Appearance Figure o. e 1. Latex stain on Dark spot -` -- surface Dry tissues beneath fright the peel or surrounding the defects 3. fungal area Brighter 4. Scratch an surface Dark E ___ 5. Anthracnose (black Dark surface with lesion) fall topology and rough surface 6. Partially Bright and dark sign dehydrated whole mango (Shriveled) .7. Bruised mango Deep dark and rough Surface S. Crack or cut on Dark and wider surface ar beneath the peel of fruit 9, Fresh mango Ligbt dark smooth surface surface 10. White or black Brighter or darker background back round 212 4,2.4. Nora-destructive quaity evaluations of mango fruits using MR. computer vistort technology 4.2,4.1. SizeSi measurements Adtia! lengt (L), width (W) and t rlckncss (1) ofnurngofrdl One of the most irnpartant parts of this study was to check the potential of near infrared based computer vision system for- the quality evaluation of mango fruit of Chausu and dashehari cultivars. For that reason, the pictures of fruits of both cultivars were acquired and cropped to 99x81 pixels (of Chars) and S1 x58 pixels (of ddsheh&ri) size to reduce the expense of whole process and computational time. Morphological features (length, width, thicliess and various shape attributes) were determined after processing (segmentation and extraction of samples from the image, etc.) of the fruits images. The complete description of algorithms for Chatrsa and dashchari fruits on LabVIEW software used in near-infra-red computer vision system} is displayed in Appendix 2d (A-2d,1 & A- d, ). As, the outer diameters (L, W, and T) among the various morphological features of fruits are very important criteria for classification and pricing of the fruits based oP the size ( ilic et at. 2007), the first effort was made for measuring of these diameters of each fruits by using image processing and caliper (manual) techniques. The descriptive statistics of the measurements displayed in Table 4.2.4. la showed that the c mputed lengths of Chausa and dathehari fruits with image processing technique were found to be varied between 50.40 mm and 107.94 mm, and 75.27 mm and 114.78 mm alongwith the standard deviation 6.43 min and 9.85 mm, respectively. The fruits of dashehari cultivar were investigated shortest as well as longest fnrit, Similarly, from the Table 4.2.4.1 a, the statistics of fruits width and thici mess measurements revealed that the dashehari fruits were little bulky in shape than Chausa fruits. In addition, very high sample variation was recorded among the IP length of dashehari fruits. In contrast, the sample variation in IP width of dashehari fruits was recorded lowest among the diameters, Furthermore, the Table 4. .A.1b also displayed comparison results of measured (V) and computed (I) diameters (L, W and TJ of mango fruits alongwith the scale factors (k) for length, width and thickness of fruits of both cultivars. This table more clearly represents the lower and higher boundaries of measurements by IP (automatic) and caliper (manual) methods along with the average value. A complete comparison between the results 213 Table 4, 4x11, jmnrt' ftaki i of ih ni' sur d of m ig' 1nii Bu adlh and'lhicLmc wren NIK CV w,rys fk~eri~4i r 1'h prep rli rnacl~ui 1!'im' proki~ a r r• .r- r Clamufruit r 1iw f it Y•. L r f..rnglh ir~rn] i r doh rrdni 1 i kn (tint) Lng1h (mm) Id a th (uim TJti ss n i Mom 95A % l 4O 97I.5' .5,8 50,73 Std F. 0.93 ,S? i160 1.3! U.45 its D x,43 3168 314 9I5 x,45 SmTl iliri 41.37 t3, I R.33 J'33 . I ?ail x].40 47.73 327 7U7 5[.2 45.0 Mix [ J7,94 52.54 57.0 114.7 (JI96 i'hp cal prpirt WA10 bI 'a i (Nnaal) 11imP.P&nN fkurlptlyeW. ddirA 1 t iii SIdudard ii'i'ii' SiiiIard NOW 5 1 ym 1ki 1phh.1. Miiiii L ~gth,nun 10002 1[~i92 B:~adth 0.55 ?. 15,3d 50.2 llil ld a:o i:i. 0,i4 4.59 edrfr Calli tr iii j.3 7.7 120,01 Bsca~tll, mug 5I l}4 51,~I 63 95 m6hr kness, mm 5,83 014 x.14 9.R4 47,27 0.13 214 'Lgble U&11~' Cfoipain of rr 'nrad (V and aumpt (W) kngth, hr th mil Ikües of iiIu 1'uiti u~cd in the Fall I nth cal r ad c~l~ Thic1ncss C~[ Cultly~r~ VC IP factor (k} VC IP fat r (k VC fc(otUk4 8I 80,41 4.$1 Ol52 4773 OI 0 33.59 33 2 OM .Max 113.91 1O7.4 6',9 -62.54 58.69 57.b Avg. 10M 95173 5 ~I9 5M3 5O4 47O ,' ]fir RSA 75277 (M. W 51.X7 51.29 (M. kJ 47.29 416 Q M x ! 2 1 114.7 f,95 13 55.70 Avg :()2 3 97 51 5718 5189 52.12 5032 TJO1k &L32 uircdt-~ t uralpis kbr friñ lmi qlt, br thh tad !hiukn s of Charua Mango cular, I L~rten lino n mmmart n tiud f v niir £li r pored (1Pj] UiniI5iofl Paired H2 SWW]ard . i a 94 rrDr a 9 CI (r thq mean '1 limit of t4cst D vl& (mm) d ffe act (min) fferiic dfflee{i m) Aprerm nt(=) T r 'h a hr Liigth 0.+0 D.95 2.Ufl 4,171 rIU9 9 159:4,f6 l3rcacltb OA fl D." Y. 1.6]75 2 4 2 .22!I3 1 ;23 P7O ;5.7 Thi~k~~ s c!11 ,9!4 1,6908 1.9659 Q, 3 1.49 ;241 ri cu1tii L mgl t D101] 31551 4,81'I 0,4119 3,I)9 " 5 66 -14 ,11.0 linedth AIRR ~. 4 1A61O 1,9276 .1*5 1.65 .2 1 -1.94 T* ii i p,DIJ r, 1,,927 k14 226; I 2 460 4.80 215 obtained from both methods is discussed below. Comparison between imugeprocessing (I?) and caliper (manual) method The measurements of fruits of both cultivars obtained by image processing method were vcri(ied at this step by comparing the results with the ones obtained by using a caliper (manual) method. The resuIts were correlated and displayed in Fig. 4.2.3.1 (a, d), (b, e) and (c, 1) for length, width and thickness of mango fruits (Chouaa, dasheh2rr), respectively, The determination coaf±icient (correlation coef cients)2 of Charna fruits were determined as 0.951 (O.975) 0.909 (0.953) and 0.910 (0.954) for length, width and thickness, respectively. Similarly, determination coefficient (correlation coefficients)2 of dashehari fruits were determined as 0.944 (0.972), 0.904 (0.951) and 0.897 (0.947) for length, width and thickness, respectively. The coefficient of determination, thus, obtained for both cultivars were found to be lower might be due to the flat shape (with two round surface) of mango fruits. The mango fruits could be placed with an angle on the fruit holder which might have caused measurements different from the ones obtained by using a caliper. Similar findings have also been reported by Kilicc et al. (2007) for the classification of beans using computer vision system equipped with color (CCD) camera. Paired t-test, the second comparison, was conducted between predicted (1P) and actual (DVC) our dirnensions (length, width and thickness). Both methods were found statistically significant (P .0) and the obtained P-values were recorded Tess than 0.01 for both cultivars. The differences between two methods are not likely due to change and are probably due to the [V manipulation. The significant difference between image analysis and the manual ( 1per) methcld, thus, might be caused by the poor reproducibility of the latter. Nevertheless, the results obtained were consistent and the correlations of the mean dimensions between both methods were noticeably high. The mean difference between length (L), width (W) and bickness (T) measured by both methods are displayed in TaNie 4.2,4.2 along with standard Jeviation, standard error of difference and 95 % CI for the rn~ean difference. The mean iif emnce between outer dimensions (L, W, TI m u red b + caliper and IP methods for 1ashehari cultivar were 4.82, 1.92 and 2.09 and for Chausa cultivar these were 4.17, 2.46 and 1.96, respectively. Further, the Bland-Altman plot (Bland & Altman, 1986, 1999, 2007) mown as difference plat, the third comparison was also conducted for compression of the -exults between the digital caliper and image processing methods are shown in Fig 4.14.2. 215 120 1 -1113 16,4 - R=0.91 1I0 100 A 14 90 0' so 60 - wO 90 100 110 120 13D 90 100 110 120 l 11rago1€1,muu(\V a d r 0, 5ixf6,431 `° 4 50 4S 45 0 [M 65 70 6i i~ra vidth.marr ( ' b 60 v ! OSS7x=3-?4l 40 a t 1x=4510 R°= b,7 Q • 444 .10 a 45 15 e SO 4 45 SO 55 bbl 65 , 40 0 ~11a oI~ic ' f Fig 4.2.4.1. Comparison of computed and measured outer dimensions (L, W, T) of Chansu (a, b, c) and dashehari (d, e, f) fruits with image processing ([P) method based on NIR CVS and digital vernier caliper (D C). to MR Camera Gz a NIR camera +1.965 +1.96 SD 8 LO 8~ 4 6 LlA 0 0 4 p a 4 8 as p 8 ry 6 O p '0 4 4 4 • C i a a O a o 4 4 L4{CLfI } Ct ¢ : a LR~-0fl T N y Y.L 4 • n4¢ O O O J1 Y i G 9 ~ 4 ~ D O 4'I] 4 8 4 4 a • ° -LSD a a L%SD 0.L -LA I I r I so W 90 95 L116 ]4 5 ] 10 L]5 121J 75 RS 95 IQ5 H5 125 ]35 AveralmIan Oi{YCM - [F) AverPigcIcnglh (MI . IPJ a d ' NI . camera s NIR Camera 4 a ° +1.96 SD +1.9C`SD 5 ° ° 4 4 5-7 a a 4.4 000 D i• ° a 4 4 OU a iJ 2 ate, o 0 0 2.5 a g ° a=0 L-9 ] I O O 4 O a 4 O 6 O 6 44 4 a ° L95 SD i 0 -1.46 SD -0-7 .0.2 -Si -1 45 50 55 60 155 70 50 52 54 56 5B 60 62 64 Av ewldlh (VC-M' - IP) Average rr(dlh (YC61 - lh) b 6 NIR camera 6 MR Camera a +1.96Sp 5 53 +I-96 SD 4.4 O • 4 O i~ 4 0 8 . 0 a ~ 3 U q 4 h p 4 4 a (p0 h o 4 8 [c~ a 4 2.0 y F 6 O 0 G `2.1 a 4 p 4 o 4 p 4 ` c ~ 8 4 0 8 4 4 0 a O P .1 .I -96+ 5L] ° .l.%Sti -2 -L3 -L AL] 45 50 55 6u s6 4S co 52 54 56 ss 60 Aver~L 004mm5 [VCM - «} Arg IhI'i~11145$ - IV) c I Fig 4.2.4.2. Bland—Altman plot for the comparison of outer dimensions (L, W, T) of uuusa (a, b, c) and do hehnri (d, e, f froir's computed with image processing (IP) method based on NIR CVS and measured with vernier caliper method (Veld); outer lines indicate the 95% Iimits of agreement and center line shows the average difference. 218 The 95% limits of agreement [(0.10; 8.30), (-0.70; 5.70) and (-1.30; 5.30) for Ghau.a and (-1.40; 11.0), (.U.20; 4.00) and (-0.60; 4.80)] for dashehari cultivar] contain 95% (3151) of the difference scares. The mean difference (bias) of the measurements [length (a, d), width (b, e) and thickness (c, ] of the fruits hausa, dashes rri) bctwen lP and VC methods were (4.Z 4.8) (2.5 1.9) and (2.0, 2.1) mm, respectively. From these results, it an be stated that mango fruits outer dimeasioi have effected significantly on the accuracy of estimated dimensions IP<0.0111). In these figures, outlier indicates the 95% limit of agreement and the centre line shows the average differences. Table 4.2.4.3. The mean values, standard deviation (S.D.) and coefficient of shape geometric attributes of mango fruits {Chausa and dashehart cultivar) analyzed by IP technique based on NIR C VS. Cultivars Parameters Minimum Maximum Mean S.D. (1) C.V. (%) cixausa cultivar Elongation Factor 1.95 2.82 2.12 0.15 7.0 Compactness Factor 0.52 0.81 0.76 0.06 0.08 Heywood Circularity Factor 1.04 1.43 1.07 0.08 7.4 Type Factor 0.75 0.99 0.97 0.05 4.9 L}ashehurz cultiver Elongation Factor U,OS 2.39 2.20 0.09 4,0 Compress Factor 0.69 0,$2 0.77 0.03 3.7 Heywood Circularity Factor 1.05 1.1 1,07 0.01 1.1 Type Factor 0.98 1.0 0.99 0.003 0.4 4.2.4.2. Estimation of shape attilbutes by image analysis Various shape attribute evaluated automatically using Lab iew image -processing software from the profile image of mango fruits acquired by NIIR canmera. Among them, some shape attributes of Emits of both cultivars are presented in Table 4.2.4.3. The mean value of fruits elongation, compactness and type factors are slightly differing bctwecn the cultivars and can be used for differentiation between them. 219 Dashehari fruits were noticed longer and also slightly flatter than Casa fruits. The Heywood circularity factor, however, of the fruits of both cultivars was same but the coefficient of variation (°%%) in factor was different, higher and lowest variations were noticed in the circularity (7,4 %) and compactness (0.08) factor for Chausa fruits. The range of variation (min & max) and standard deviation in shape attributes can also be noted fr+oin the Table 4.2.4.3, In addition the other shape attributes were also evaluated automatically by software and cocrelated to the fruits weight and volume. Table 4.2.4.4. Coefficient of determination (Ri) and linear regression weight models on the basis of area (pixels) obtained by image processing technique based on NIR. CVS. . Shape attributes Regression model R-square R-square RSE (Regression {pixel} (adjusted) Standard Error) Chausa cultivar Area y = 0.O88x - 3.37 0.835 0.832 12.66 Perimeter y 2.189x - 250.4 0,712 0.707 1634 Max Feret's Diameter y 5.338x - 228.5 0.701 0.696 17.06 Waddel Disk Diameter y =7.153 x - 243.6 0,740 0,736 15.89 Daskehari cultivar Area y 0.050x + 46.04 0.756 0.751 11.47 Perimeter y= 1.241x - 62.19 0.697 0.691 12.79 Max Feret's Diameter y _ 2,578x - 21.13 0.703 0,698 12.64 Waddel Disk Diameter y _ 4,434x - 81.75 0.72-9 0,729 12.09 4.2.4.3. Relationship between fruit's shape features and weight Linear regression model was used to develop the relationships between the shape features and weight of mango fits. Table 4.2.4.4 displayed the coefficient of determination {R2) RSE and linear regression weight models, for Chausa and dashehari cultivar separately, based on area, perimeter, Max )Fret's and Waddel disk diameters of fruits computed with image processing technique. 220 254 VZQ. X-63x,, ?3Q -~ •r = 0.05i 4.04 ~" 20G~U935 D i5u 10 4 loa 100 `. 50 50 1300 2000 2 D 3M 1000 15fl llJ ~ 00 30O 3500 { :1Et~ ~i ixele~ al 23 Y=14l•6w1 230 y=3,189x-150.1 ;50 R2=71, 190 ~ 1#Q 100 do 54 _... l'-o 140 160 18O :Qo ?'0 5Q —~ Pttitti uff~aiL (rL W 100 150 : 0 :S P r tieheroffni;ts{ ae13} bi b - s=5,335-,23.9 R`~ 4,701 ? R=- 0.703 ,OO 1 o ~+F 0 100 100 40 50 GD 70 80 9] M 50 60 14 S 90 1axFerel diaa 6a- (j ixel~) I41 Feet dime i (pielx) [l- C2 1_[]'_G =dkr43•~a-81,75 7 +'-7.113%-343.6 R"= 0.7'_4 1ST I 140 4} v s0 1 ______4D Me 6I1 40 5 Go 7a WaddeldAdlaDet f iKels) Taddeldi 1:diao~octar yi e1) d1 dz Fig 4.2.4.3. Correlations between shape attributes [area (a1, a2), Perimeter (b1, b2), Max Feret's (C], e) and Waddel Disk (di, d) diatnetersj of Chrm.sa (1) and dashe/iari (2) fruits computed with particle analysis by image processing (IF) method based on NIR CVS and weight of fruits measured by digital electronic balance (DEB). 221 The R-square value pied between 0.701 and 0.835 for C'hausa fruits while for dashehari Emits it was varied between 0.697 and 0.756. Between surface area and weight of Chausa fruits, highest coe# ciert of determination was obtained with lower RSE in estimate. lowever, the RSE in the estimation of weight on the basis of shape attributes of dashehari fruits were lowest. The single side area of fruit, obtained from the image acquired at the 0° arientatioiilposture, was considered for the correlation [between area (pixel) and weight (g) and area (pixel) and volume (ml)] study in this research. However, the areas in other posture (0°, 90, 180 , 270', 360°, top and bottom) of fruits' 'ere also evaluated but not entertained and reported in this research because of overlapping of one area in more the one postures. 4.2.414. Relationship between fruit's shape features mod volume The outer surface area of mango fruits obtained from the image acquired at 00 posture was found to be having higher relationship with the volmne of fruits. As the R-square value was obtained >0.82 and >D,744 in case Chausa and dashehari etilLivars, so by measuring area, the volume of fruits can be estimated precisely. Similarly, as the R-square value for each shape attributes were predictable. Table 4.2.4.5. Coefficient of determination (R2) and linear regression volume models on the basis of area (pixels) obtained by image processing technique based on N[R CVS. Shape attributes Regression model R-square R-square SE (Regression (pixel) ( adjusted) Standard Error) Chausa cultivar Area y = 0.085x - 60.90 0.823 0.820 12.70 Perimeter y = 2.035x - 227.1 0.56 (1.650 17.74 Max ?eret's Diameter y ' 4.92x - 208.5 0.650 0.644 17.88 Waddel Disk Diameter y — 6.S 9x - 233.6 0.727 0.722 15.79 shehari cultivar Area y 0.O4&x + 46.73 0.749 4.744 11.20 Perimeter y= 1.175x-54.40 0.675 0.669 12.73 Max Feret's Diameter y = 2.481x - 18.63 0.704 0,699 12.14 Waddel Disk Diameter y = 4,71 lx - 73.67 0.711 0.705 12.01 222 20 1 251 Y=0.E Sx• 6090 Mo ] R'= 0.123 R= 0749 200 100 0 0: 1000 1500 3000 :500 300 3500 Area (p gels} 1540 2000 area ~;x.)?504 30IG0 a1 a2 1t - 0,050 ti 1SO ]UD - 54 no 11J 160 130 2200 224 :00 150 200 250 Payuwt~r Offivits (pixel;) Puimneterol'fi a (p Pk) b2 50 ° 3431x • 18.63 R'=O2704 R Ct 50 0- • r 150 i ai 10Oi 100 40 50 60 70 80 90 100 ~C d0 74 $0 90 ;riai i et din (pied) MixFaet 4[ i ti {p r]s) C] C2 250 ? 230 ] ti' =4 21lx- 73 67 ; R'= 0.7?? • r ;04 ; 154 - 100- 100 4D 5D Q 0 40 50 G0 70 Taddeldiftdimnere~'(pixeis} W mddeldiskJimnder~pi 1s1 d, di Fig 4.2.4.4. Corrc]ations between shape attributes [area (a,. a2), Perimeter (b1, b2), Max Feret's (c', c7) and Waddel Disk (dt, d2) diameters] of Chausa (1) and dashthari (2) fruits computed with particle analysis by image processing (fl) method based on NIR CVS and volume of fruits measured by water displacement method. 223 However, lower value of R-squared suggests that the variation in the volume of fruits less dependent on the shape attributes evaluated by IP method. 4.2.4.5. GeometricalGee attribute based modeling of fruit weight In order to estimate weight and volume of fruits using image processing technique, CVS based on N!R image acquisition system can also be used. On the basis of geometrical attributes eactractei from NiB. mango image profile, £ourtoen models were suggested for estimation of fruits weight and volume of Chausa end dasheJiari cultivars (Table 4.246a b) Fruits from both cultivars were considered in combination for developing of these models. Motteling fruit's is ass The R value of alibrationlvalidation regression equation for weight was noticed between 0,634!0.616 and 0.932/0.92-6, respectively. The lowest cccfficient of correlation was obtained for the thickness or minor diameter (model no. 3) of mango fruits extracted from the profile image acquired by AFAR camera. This suggests that only 63% variation in fruits weight can be explained by the model no I In contnst, length (major diameter) of fruits has good correlation (0.81) and width (intermediate diameter) of fruits has reasonable c lTelatinn with the weight. The maximum iorrelation coefficient was found for the model no. 8, 10, 11, 12, 13 and 14 while minimum was found for the regression equation based on thickness of the fruits. Further, the model ono. 8 can be selected as best model on the basis of higher R value (0.947), lower EC/ EP value (10,68/10.93), lowest difference between SEC and SEP (0,25) and use of single variable (Da) in regression equation. The model no. 13 can also be selected as one of best model on the basis on highest R value (0.932), lower EC/ EP value (10.66/11.04) and lower difference between SEC and SEP (0.28). But, this model (nn.13) is more complex than the model no. 8 because this one is based on the length and geometric mean diameters of fruits. In other words, model no. 13 is based on all three diameters and measurement of all these is neeEed which makes sizing rnecJ nism complex, tedious, costly and time consuming, Similarly, model no, l0 and 11 can also be useful for prediction of weight of the fruits of Chausa and dashehari cultivars. Except model no 3, all models were found to be reasonable fit of data. 224 TahIe 4. ,4. , Vadeng MLR model; bi ed u f dim si I churn tk {]ngth, breadth 1J1 $Icbc s) fur dctnni.air wvi lid of t«lilgu tiviI Df Ctt te duMol Ou1tirEt'. S. Madds R;1hri n SECtEP Rlns CorreIR1ii Nii dibilou V4 l~ M ~l 166i' V EhbR CHIR `6 Vaud i i 14.7 1117 , ,4 0.016 II1 p i= :~~+ -S.i -t 8~'~ 1L3) j24 ,416 '&022 637 O16 4 NI - K' E '+A 9 -3,1b 1 L1~#4 1124 !1S3 1I X filXl9 O76 O.69 3 tot=, ~L Y, ad=3. i 1 -' ~~ F . 4 1.43 1172 - IA J0 •OIC )9 OI 72 O6 6 . ILWE X -4,92 LO4L Ti4 3 1],99 122 0U 0, 06 4881 OI,S75 +K2 ~-2 i.4i 1107 1237 -11t .U17 0,$7 1)$73 3 i:: KID,+T2 1+1 -0Q 9U7 019 1= ,t+ jr+Xa M~3.2DL+2,99-I .4] 11i2 11.92 4.OG0 -0,416 0691 0,$$ 14 M. K1T1K1 V,KT+K4 Mr]7T,tl.72w+:7ST 19L.r 10,64 1I.] • 1 I113 IOS 0.598 11 = IL+ r+ } M=I. F+3 xIF, —t i.l 10.59 IO,9 4ag44 -01 , ]3 M= If•I./ L TH3 . Mil .E3.49X,14L('-IU 9 10.6 1I4 , • ,fl17 019 7 0.00 13 iR. IE+x2D tEJ .u=1.0uf X -IS 11L66 111)4 O.ON -DM3 0.9.32 0.926 14 -. ,r.1 ~~~~ ~ u- .az~+$, Ira —1%61 1064 1102 D4 -D14 cF.9S 0.91 Various MLR mo&1 Dosed on IF dim clonal ràip (I~ng1i, it ulth and Nbc) focje mining vc~Iiiim of mango frni1 1Cka, md h ri uuIlivars. r - - 1rt-- 7LLi! Rt IOIi 17hi WY ~LL C.,AWh n Vri]In Iiltr1i V~]Ià Q11 cslliair F81rMon 1 V = L4 M1 V = J. r 7.9 l.13 I9. O. O14 ? 0, 5.331-I 1 217 2,73 [x,1101 O. i6 v=Kr+x; V-I. 4;'`+ '1 2IY7 1i5 t.DLk 4.412 x.614 O.6Li4 , + r7 16.12 0.000 0013 01805 4,792 v= 'i r Y=3~a ;a ~ ~ + ~. 18 7 18.4 .O.88 O' 11]22 4. 'K1 Li7K1 v=a~~ aa41~~ +59.az 1L145 17.94 a,OOl Qr0f1 1749 f73 v +K. V.5Ii1.1 I+10 IN .01 (]857 x.761 0.746 VD +K V 4.-I46 1756 1Bd IUO 0,475 11745 4,121 v=I,i+K J'+ , v•14?L.w~.1 T-IIIA 17.72 1831 4MC?O -ORO f1,740 0.712 14 VmK,L*j2W+k;T+Kd vU+1 '+ .4 '-l~9i 17.G 1B ORG 0,411 7 O .11 1 r =K,L+]C~ 'f'+fir, V•1.2IL+4,21 14~9 '— 4 17. 17 -0 UO -OI01 01761 OJ73~ 1 17.4 17.2 J1OJ1 U P U 4.'21 13 1r=K,G+xtD,.1ka =1.14u:p~~.+s.~~ -«o.s 17,10 11.$8 .0,O p 4,003 0161 0,733 14 V41 L+K+K3 v=-1,~ a 1. jo-i'?.~R 176 175 Ir11O4 r11I4 12 .P 4 Modeling of mango fruit's volume Table 4,2.4.6b shows the models suggested for volume estimation. Model no. 4 had highest R value (0.805) and lowest value of SE EP (15,67116.1.2) can be selected as best model on the basis of strong correlation. As this model is based on multiplication of two variables, i.e. IP length (L) and width (W), the two diameters of fruits must be evaluated for the estimation of fruits volume. In contrast, model no. 1, 2 & 3 based on single diameter of fruits (L, W and T) were simple, cost saving and time efficient but had slightly low degree of coefficient of correlation. In addition, model no. 7, 10, 11, 13 and 14 have the R value more than 035 indicate that correlation strength is the reasonable and positive. Further, the correlation coefficient of calibration/validation regression equations for volume estimation varied from 0.624/0.604 to 0.805 to 0.792 with no sigrffi ant bias in calibration and validation, respectively. Similarly, from the Table 4.2.4.6b, it is evident that the models based more than one diameter or variables of fruits are having greater degree of relationship than the single. Further, the fruits AMD GMID, and aspect ratio, sphericity and total surface area of mango fruits (Chair a and dathehari cultivar) evaluated by IP techniques using near-infra-red computer vision system were also tested during the weight and volume modeling. But, these parameters were found weakly correlated to the mass and volume. The model based on these parameters, hence, not reported in this thesis, However, the mean values, standard deviation (S.D.) and range (ruin,, max) in real world have been reported in Appendix 3 (A-3,4). For the total surface area (real world: mrn2), the area obtained from the image acquired at the 00 and 18V0 orientations/posture.s were only coisidered. 4.2.4.6. Defect detection of mango fruits using NIR computer vision technique The results from the image profiles acquired at different wavelength bands such as 730 rim, 766nm, 780nm, 800nm, 830 nm, 852nm, 855nm, X80 nm, 905 nm, 940 nm, 1000 non 1050 tiro, 1064 mn, l lOOnm, 1150 nm, 1206 urn, 125Onm, 1300nm, 135Onm, 1400nm, 1450 urn, 1500 nm, 1550 nm, 1600 nm and 1650 nm (Fig 4.2.4.5) suggest that the three wavelength bands 1450nm, 1500am and 1550nm in NIR based computer vision can he used for the detection ofbraise or defect underneath of the peel of mango and defected tissues. 227 In contrast, Lu (2003) has reported that the spectral region between 1000 nm and 1340 nm is most appropriate for bruise detection of Red Delicious and (lolden Delicious apples. This finding might be differed t0 the difference in commodity. Maichow (2007) has confirmed that using NIR-responding cameras, imaging beyond 1000 nm improves the capability to see transparent coatings and reported that fruit and vegetable inspections with NIR cameras can detect and reject bruised fruit. Similarly, Tao and Wen (2003) also reported that the near- in r=di (NIR) range from about 750 urn to 1200 rim is senaitive to detect defects on fruits but separation of true defects from stem-end, stem, and/or calyx using only near-infrared is very difficult, To scLve this problem, a novel method was developed which incorporates a near- infrarud (NIR) camera and a mid-infrared (MIR) camera far simultaneous imaging of the fruit being inspecteii The NIR camera is sensitive to both the stem-end.icalyx and true defects; while the M R camera (5000 urn to 8000 nm) is only sensitive to the stern-end and calyx. 'roue defects can be quickly and reliably extracted by logical comparison between the processed NIR and MR images. Infrared camera color images (all gray level redlgreen) are converted to black and white images which are then compared and contrasted to one another to subtract out the non defects such as stem-end, stem, and calyx to leave only true defects. Bruising normally happens to the tissue beneath the foot skin. Aflar the fniit tissue is damaged, its cells are initially filled with water and turn brownish. As, time elapses, the damaged calls stark to loss moisture and eventually became desiccated and blackish, This type of black spot, if any, underneath the peel of mango, using NIR image acquisition with wavelength 1550 nm band pass filter can be detected. The strongest response, in this research, at the band pass filter that enhanced bat the intensity contrast between bruised and unbruised mangoes observed for 1550 um differed from the finding (750 nm to 1200 nm) reported by Tao and Veen t2 O3) for apple, it might be due produce difference. In addition, the areas of fruit having bruised tissues appeared darker than th3 - unbruised trines after capturing the images using NIR. acquisition system might be due to absorption of infrared light by the water released from cells when IIeshftissues are damaged/bruised. 228 940 nm nl ] ~otvn I 90512 13y 5~ I c5O 111 am nrn nm nm f101)nm 1 I2 12 r, i 5l m 1~IJJnm 55i 4'ithnut Tiler F. 42.4.5. his of mango fruit uii4 t ii :lg N!1 imagc acquisitinr. R}itcir. tnnii i rli rent w~vdcngth (nm) filter A Z29 The mango fruits skin is generally not broken because of daruagelbruising of the surface during the harvesting and handling. The cells in fruits flesh are damaged and nutrient along with water in the cells are released and fills the spaces between the cells. The infrared light energy penetrates the skin into the flesh when fruits are illuminated and consequently, their images were captured using NIR camera. In this case, there are no tiny gaps to diffuse the infrared and thus water absorbs the infrared light. The bruises on fruit's surface appear darker than the non defective parts of surfice. The black spot underneath the peel of mango how look in the cola (RGB) image, in the NIR image (when taken at I SOnm wavelength) and when thresholded and peeled off is given in the Fig 4.2.4,6. For the quantization or calculation of percent area of black spot underneath of peel in terms of defected area, an algorithm is developed (Fig 3.21.) and discussed in Chapter 3. With the help of developed algorithm, we can also count the number of pixels comprising the entire fruits surface after thresholding. Consequently, the ratio of low gray level value pixels to the total number of pixels can also be calculated. A high ratio means that there is a big fraction of Iary gray level pixels which can be suspected to be of bruised surface. The ratio can also be compared against another specified threshold value. If the ratio is less than the threshold, we can level the mango fruits as good, otherwise are bruised. Thus, the NIR acquisition system using band filter 1550 rim waveiength can be used for the detection and characterization of defect underneath the peel of mango on the basis of appearance. The appearance of bruised and unbruised cell underneath the peel of mango was bright and dark, respectively while the appearance of water core and fungal organism just below the peel of mango was darker. A similar result of NIR imaging was also has been reported by Uthaisombut (1996) for cherry fruit. Unay and Gosselin (2007) used different visible and NIR interferometric filters to recognize stern ends and calyxes in sound and defective Jonagold apples and reported that N[R filters were overcome the problem posed by the fact that this cultivar has a variegated red and yellow surface that males it more difficult to detect skin damage by using standard color images. 230 CcI NIR image Image after Color operator Operation In TI O. Image aft off Fig. 4.2.4.6 Appearance of defected area on fruit's surface in color image. NIR image, after thresholding and after peeling. 231 Similarly, AMMi =s et a]. (2002) have presented a multip.ectral camera system capable of acquiring visible and NIR images from the seine scene and used specific algrrithms that were implemented on two processors and worked in parallel. They found system capable of correctly classifying lemons and mandarins at a rate of up to 10 fruits per second, depending on the area of the external defects oLnd the size and color of the fruit. Xing et al, (2006) also identified several visible and NIR wavelengths that could be used to discriminate the stem end and calyx from sound peel and bruises in apples. 232 Chapter 5. Summary, conclusion and Recommendations for future work The summary of main results and principal conclusion drawn from the current research has been included in this chapter. This chapter also includes the areas of significant contributions to knowledge together with possible areas for future investigations and research, The section 5.1 summarizes the main findings of the research. The section 5.2 discussed the main conclusions reviewed in terms of experimental results, feature extraction and performance of the manual (destructive. method) and image processing (non-destructive method: using four cameras such as monochrome; color, LIV and NIR and two different lighting i.e. visible and UV setup) techniques for quality evaluation and classifications of mango fruits, possible areas of future work are presented in the section 5.3. 5.1. Summary The sorting and grading of mango fruits according to fruits size, shape and surface defects, which govern the quality standards and markets prices, is very important task before exporting in national as well as global markets. These quality standards of mango fruit generally supports for high commercial value in national and international fruit markets and likeness by peoples. Quick, accurate and efficient grading requires automatic and high degree of sensing and intelligence for accomplishing the sorting and grading of fruits. In India, manual inspection based visual approximation which is tedious, laborious, and costly, etc., is often used to control the quality standard exporting the flits, but no automation based on non-destructive techidques such as computer vision system, etc. have so for been used for inspection of mango fruits. Before sending consignment in global markets, fruit exporting bodies (government as well as private) are seeking to modernize their fruits quality grading system in novel and scientific way to satisfy a market driven by local as well as global customer demands with the fruits high differentiation. Consequently, the research on real time quality analysis or sorting/ ailing of fruits lies gain momentum so far. A scientific and novel approach by using computer vision system, one of them (iowdestcuctive techniques), for sertiuglgradi g of mango flits on the basis of surface parameters is carried out in this research. In addition, the effects of the various factors such as 233 harvest stage, cultivar, storage period and their all two way interaction CO physico-biuchemieal parameters (length, width, thilces, AMD, UMA sphericity, aspect ratio, specific gravity, total soluble solids, titratable acidity, total caxotenoids and colour values) of mango fruits were studied in the year 2010 using the subjective (destructive; manually) method, in this research. Four cultivars. Qwausa, dashehari, Langra and Malda were included for the study of variability in quality of mango fruits that governs the market prices. However, two cultivars i.e. Clumsa and doshehari were included for the non-destructive quality evaluation by using computer vision system, in year 2011. The minimum, maximum, standard deviation, sample variation and average values of mango fruits length, width and thickness were nieasuracl with digital vernier caliper (DV) and computed with image processing (IP) method after capturing the fruits images by using the different cameras such as monochrome, color (RGB), ultra violet and NIL and lighting (UV and visible regiwes) were compared by using methods such as regression analysis, paired t-test and the Oland—Alt van plots, The scale factors (k) for length, width aid thickness of Wits, when image acquired by using different cameras, of Ghausa and daseh.ari fruits, v~'ere evaluated in this research using two shape i.e. rectangle and square and then average were taken of all measurements, respectively. Scale factors were obtained by rationing the measured dimension in pixel and actual dimension in mm of perfect rectangle of wooden shape to convert units from pixel to mm. The results computed with image processing (IP) and digital vernier caliper (DVQ method were compared and found that the coefficient of detemi vatic n (112) for fruits length of CJu sa and das/ ehari cultivars when fruits image captured using the different cameras such as monochrome, color (RUB), UV and NIR were 0.948, 0.920, 0.907, 0.897 and 0.935, 0.904, 0.885, 0.825, respectively and for fruits width, the coef cients of determination (R) were recorded as 0.922, 0.874, 0.817, 0.826 and 0.917, 0.877, 0.800, 0.781, respectively. Similarly, the coefficients of determination (R2) for fruits thickness were found to be 0,909, 0.S5, 0.S20, 0.828 and 0.891, 0.863, 0.834, 0.804, respectively. The coefficient of deterziination (R2) values can be interpreted as the propnrtidn of the variance in the IP estimates attributable to the variance in the actual measurements. The higher the B.2 values, the closer the IP results arc to the actual (DVC) results. The proposed [P method yielded above 82%, 07S % and 80% accuracy in 234 estimating mango fruits length, width and thickness, respectively for all camera system. The, maximum accuracy was observed when monochrome camera was used for image capturing of fruits as 94,S%, 92,E% and 90.9 for length, width and thickness, respectively of Chausa cultivar while for daghehari cuitivar, 93.5 %, 91.7% and S9.1% accuracy for length, width and thickness, respectively were observed by IP method. The second comparison was provided by means of the paired t-te t results. The mean dimension of fruits difference between IP method and the standard deviation (ad) of the dimension (length, width and thickness) of both cultivars differences have been reported in this research. The obtained P-values for length, width and thickness of fruits of Chauca were 11.22, 10.82 and 7.86, respectively and P-values for length, width and thickness of fruits of das/shad 11.23, 7.40 and 16.731, respectively when monochrome camera was used for image capturing of fruits. Therefore, the paired t-test results confirmed the dimension computed with IP method and DVC are statistically significant at the 5% level (P > 0.05). Further, to reveal a reIat unship between the differences and the averages, to look for any systematic biases and to identify possible outliers, the Bland-Altman plot or difference plot was also used and found that the mango fruits diameters have no effect on the accuracy of estimated dimensions (P>0,O5). This plot indicated that the diameters differences between the computed and actual diameters were normally distributed. The outliers indicate the 95 % limit of agreement and the centre line shows the average differences. Furthermore, using simple mathematical expressions relating simple fruit global dimensions is an approach to describe fruit shape attributes. The measurements such as area, perimeter, Max Ferets and Waddel Disk diameters, and elongation, compactness and type factor obtained from the particle analysis of fruit's image, from all camera and lighting setup using Lab View software. In this study, it revealed that the fruits elongation, compactness, type factor and Heywood circularity factor can be used to differentiate the fruits of the both cultivars. In contrast, fruits area, perimeter, Max Feret's and Waddel Disk diameters were correlated with the fruits mass and volume. In addition, major diameter (length), intermediate diameter (wir!tlr) and minor diameters (thickness) and various derivative parameters (such as aspect ratio, sphericity, arithmetic and geometric diameter and parameters generated by different combination of length width and thickness of fruits, etc) based on these diameters were also modeled for the estimation 235 of fruits weiightlmass and volume. For the weight and volume estimation of mango fruits, models suggested for monochrome, color (ROB). U'V and NIR camera setup were 20, 13, 15 and 14 respectively based on geometrical characteristics of Emits computed by using IP method. Models based n more than one variables have higher coefficient of corre ation with weight and volume of fruit for all camera and lighting set-up except in case of I TIR camera for weight/mass estimation model which dependent on single variable i.e. length. In contrast to the models based on more than one dimensions which makes sizing mechanism more complex and expensive, the model based on single variable (i.e. length in case of other camera and lighting set-up) can be considered best model on the basis of simple sizing mechanism with respect to cost feasibility, and good coefficient of correlation, The higher coefl5icient of correlations was obtained for the models based geometric attributes computed with IP method in case of monochrome camera (0.671 to 0.934) followed by IJY and NW camera (0.624 to 0.932) and then color camera set-up. The models based on aspect ratio and sphericity (for all cameras and lighting regime setup) was correlated inadequately with the mass and volume of fruits, hence the same was not reported. The maximum elongation factor was recorded for fruits of dzrkehar-i (i.e, 22l. 2.23, 2,36 & 2,20) than the, Chausa cultivar (i.e. 2.10, 2.11, 2.23 & 2.12) in case of monochrome, color, UV and NIB. camera setup, respectively. Except in ewe of UV camera setup, the values of elongation factor for all other camera setup for a particular cultivar was almost same. The dashehari fruits were found more elongated in shape than the Ghausü fruits. Similarly, the compactness and type factor of fruits varied for bath cultivers and camera setup, The value of compactness factor for Chausa was found 0.76 shows circle indentation inside of the circumference while in case of dasheiwri the, value compantness .7 , except in rye of 41R camera) shows circle protrusions outside of the circumference. However, the Heywood circularity factor, another well-known particle measurement used for shape analysis evaluates analytically an image's border and found almost same for both cultivars when image of fruits was captured using monochrome, NIR and UV cameras. In case monochrome and NIR camera setup for image acquisition, the Heywood circularity factor was 1.07 while in case of UV camera it was 1,09, which show that the shape of fruits are closer to a disk. However, licywood circularity factor differed for both cultivars when images of fruits were captured using colour camera setup. So. the Heywood circularity can used to classify fruits of different (these) cultivar by using color sera. In spite of above, algorithms for defects detection on mango fruits surface were developed using morwoh one and colour camera. Simultaaneousl , the performance of developed algorithm was also evaluated on the basis of accuracy, efficiency and average inspection time. The performance of developed algorithm for monochrome camera was better than the performance of the algorithm developed for color camera. The accuracy and efficiency obtained for monochrome and color camera was 88.75% and 88. 6%, and 97. % and 93.3%, respectively. Similarly, an algorithm was developed using NIR camera for bruise detection underneath of the mango wits skin, This study also indicates that the three wavelength bands, 1450 nn 1500 nm and .550 nun, in N[E image acquisition can be used for the detection of bruise or defect underneath of the peel of mango and defected tissues. Conscuently, the appearance of bruised and unbruised cell underneath the peel of mango was found bright and dark, respectively, while the appearance of ratter core and fungal organism just below the peel of mango was found to be darker. In contrast to monochrome, color and MR, the ultra-violet camera was used to detect various defects such as bla& lesion., mechanical damage, latex stains and shriveled fruits in ultra-violet Light. The band pass filter of 400 nm wavelength was found more suitable to detect such surface defects or bruises of mango fruits. 5.2 Conclusion A novel approach for feature extraction as well as defect detection on the fruits surface and just below the peel of mango fruits has been proposed in this research. The features information obtained in manual analysis were also conjugated along with the information obtained by non- destructive analysis, separately, using each carnera setup among the four acquisition system. All these four image acquisition system are successfully implemented for both quality features extraction and defect detection of fruits of both cultivar. This new technique of shape feature/attributes extraction is found to be successful in differentiating the images of fruits from the one cultivar to other. For instance, elongation factor of dashe ar i fruits found to higher by all camera and lighting setup than Chausa fruits. On the other hand, in the experiments of various types of defects such as anthracn.ose (black lesion), latex stain, mechanical damage, water core and black spot underneath of the peel, and shriveIing nature of fruits detection using this method is found to be more successful in comparison to the traditional method, The accuracy of Z37 algorithm developed for defects detection using monochrome and color camera was obtained 88.75 % and 88.6 %, respectively. The efficiency of algorithm i.e. 97.88 % in case of monochrome camera setup was found to be higher than the efficiency of algorithm in case of color camera i.e, 93.3 %. In addition, lower average time for inspection was found for monochrome camera setup in comparison to color camera setup. Although, the average time for inspection found to be increased as the severity of defects on mango fruits surface increased. Furthermore, mass and volume modeling, respectively on the basis of the major diameter (single variable) and all three (major, intermediate and minor) diameters (multiple variables) were identified as the best models. In addition, for the better results the mass and volume were also correlated with the morphological features (area, perimeter, Max Feret's and addel disk diameter) of the mango fruits and found to be highly correlated with area (pixel) and addel disk diameter (pixel) of fruit. Finally, in context of the overall objective of the research, computer vision (non- destructive) technique is explored as a more successful technique than the manual (destructive) technique in detecting the mango fruits quality that govern the market prices. 5.31 Recommendations for future work This thesis report has been completed in delivering destructive (manual or subjective) and non-destructive (computer vision - real time image processing) technique to assess mango fruit quality for sorting{grading process. The potential of four different image acquisition systems in the Computer vision was examined. In this contest, different quality parameters of mango fruits were efficiently exhorted in terms of morphological features and bruise on surface and underneath of peel by using four different fruits image profile capturing techniques. This research, nevertheless, can be further extended in the following manner: In the Chapter 1 and Chapter 2, as stated that the mango fruit monitoring during sorting/grading is a tedious, laborious, and costly; and easily influenced by physiological factors, inducing subjective and inconsistent evaluation results, However, several different advanced techniques are being implemented for the sake of real time sorting/grading of fruits, so far. This report has documented the computer vision as useful 238 on line technique for such (rapidly, economically, consistently and even more accurately and for objective inspection) purposes. Some other parameters such as color, pixel intensity, texture, etc., which are not considered in this research, therefore, they can also be considered for evaluating and correlating to the other fi its qualities using some efficient techniques. For instance, the pixel intensity can be correlated to the bia-chemical parameters as well as to the freshness quality of fruits. The image acquisition setup can be modifcd more carefully so that the findings from actual and image processing method differ insignificantly aIong with the, high degree of correlation and consistent results. Sc,rne other morphological features such as: moment of inertia, convex hull features, equivalent ellipse feature, etc. can also be tested for differentiation of the cultivars. The flavor or aroma prvfiIing of rnango fruits for sortingfgrading was not included in this thesis. Therefore flavor or aroma which is an important quality of fruits characteristics in terns of judging the taste can also be analyzed using electronic nose, electronic eye, etc. Similarly, I IR spectroscopy, hyper-spectral method, thermal imaging, etc. can also be applied for the efficient sorting/grading of mango fruits. 239 References Abbasgholipour M1 Omid M, Keyhani A, Mohtasehi 55 (2011) Color Image Segmontatinn with Genetic Algorithm in a Raisin Sorting System based on Machine Vision in Variable Conditions. Journal of Expert Systems with Applications 38(4): 3671_3678. Abdullah MZ, Abdul-Aziz S, Dos-Mohamed AM (2000) Quality inspection of bakery products using color based machine vision system. J FoodQua! 23: 39-50. Abutaleb AS (1989) Automatic chresholding of gray level pictures using Iwo- dimensional entropy. Cumput. VisiuziGrupliics Image Pro cess. 47: 22-32. Agcrskans J (1975) Thermal imaging-A technical review. In Proceedings of Temperature Measurement, London. April 9-11. 375-388. Ahmad IS, Reid JF, Paulsen MR. Sinclair JB (1999) Color classifier for symptomatic soybean seeds using image processing. Plant Disease 83: 320-327. Ahmadi E, Ghassemzadeh HR, Sadeghi M, Maghaddam M, Neshat SZ (2010) The effect of impact and fruit properties on the bruising of peach. J Food Engi 97(1): 110- 117. Akar R, Aydin C (2005) Some physical properties of gumbo fruit varieties. Journal of Food Engi 66: 387-393. Alasalvar C, Al-Farsi M, Quantick PC, Sliahidi F, Wiktorowicz R (2005) Effect of chill stoi' ge and modified atmosphere packaging(MAP) on antioxidant activity, anthocyanins, carotenoids, phenolics and sensory quality of ready-toent shredded orange and purple carrots. Food Chem 89: 69-76. Aleixos N, Blasco J, Molto-E (1999) Design of a vision system for real-time inspection of oranges. Pattern Recogn Image Anal 1: 387-394. Aleixos N, Blasco J, Navarron F, Molto E (2002) Multispectral inspection of citrus in real-time using machine vision and digital signal processors. Comp and Flectr Agric 33(2):121-137. Alfantiii Mai, Sheriff ARM, Shafri H I , Sawed 0M8, Eshanta OM (2008) Oil palm fruit bunch- grading system using reel, green and blue digital number. J Appi Sci 8(8): 1444--1452. Amadasun M, King R (1989) Textural features corresponding to textural properties. IEEE Trans on Syt Man and Cybern 19(5) 1264-1274. Amin M, Malik AU, Khan AS, Javed N (2011) Potential of Fungicides and Plant Activator for Postharvest Disease Management in Mangoes. Int. J. Agric. Biol., 13: 671-676. Annamalai P, Lee WS, Burks TF (2004) Color vision system for estimating citrus yield in real time. An AS AE presentation paper, paper No. 043054 240 Anon (2006) ww.ficciagroindia.com/postharvestmgmt/frmits/mango.htm (accessed date 15 Ian 2006) Anon (2010) Newsletters, http:I/www.icar,org.in/ accessed date: 10.5.2011). Anon (2012) 0 & M Newsletter. Maine Department of Environmental Protection, Bureau o. ` Land and Water Quality. rnaine. v/de l atrr/newsletterf mneivs/febru srv2D12 (accessed date 12 September 2012). Anonymous (1996) Image processing and image enhancement. t:/v.ttie.aps.org/v.tkieaps.org/ • (accessed date 23.2.11) Anonymous (2003) Mango: postharvest care and market preparation. Ministry of Fisheries, Crops and Livestock, the Now Guyana Marketing Corporation (NGM) and the National Agricultural Research Institute (NARI), Postharvest Handling Technical bulletin No. 3. nynnous (2007) Guidelines for export of ludian tuangoes to USA Agricultural1 & Processed Food Products Export Development Authority, www.apeda.gov.in (aocesed date 03.02.2014) Anonymous (2009) Computer vision. http:I/en.wikipedia.ora/%vilci/Computer vision (accessed date 25/2/2011). AOAC (1994) Official Methods of Analysis. Association of Official Analytical Chemists. 1111 North 19th Street, Suite 20, 16th Edi. Arlington, Virginia, USA. 22209 raiza E, Osuna T, Siller j, Contreras L, Sanchez E (2005) Postharvest quality and shelf- life of mango culti'ars grown at Sinaloa, Mexico. Acta Horticulturae 682, 1275- 1279. Arauz LF (20 00) Mango aathnienose: economic impact and current options for integrated management. Plant Disease S4:60(]-61 1 - Ariana DP, Ln R (2008) Quality evaluation of pickling cucumbers using hyperspectral reflectance and transmittance imaging: Part I. Development of a prototype. Sens lnstramen Food Qual 2:144-1511 e ud- inas MR (1995) Evaluation of plastic films for packing intermediate moisture content of mango. Harvest and postharvest technologies for fresh fruits and vegetables of p. 306-312. Atkin OK, Tjoelker MG (2003) Thermal acclimation and the 'dynamic response of plant respiration to temperature, Trends Plant Science. 8: 343-351. Aviara NA, Shittu SK, Haque MA (2007) Physical propordes of guna fruits relevant in bulk handling and mechanical processing. Int Agrophysics 21: 7-16. 241 Aydin C, Ozcan MM (2007) Determination of nutritional and physical properties of myrtle (Mynas o!nrm O, L.) fruit growing wild in Turkey. 3 Food Engi 79: 453-458 Baetea 1, Donn K. Boedrij S. BecketsW, Clausen E (2007) Autonomous Fruit Picking Machine:A Robotic Apple Harvester♦ 6th International Conference on Field and Service Robotics - FSR 2007, Chamonix : France. Baez-Sanudo lt, Bringas-Taddei E. Rodriguez-Felix A (1999) Quality standard grades for Mexican mangoes and application methodology. Journal of Applied Horticulture, 1.5-10. Bagade 85, handily VE (2011) Use of histogram equalization in image proeesing for image enhancement. Int Journal of Software Engineering Research and Practices 1(2):2231-2048. Bagshaw J (1989} Mango pests and disorders. Queensland Department of Primary Industries, 44p. (Bulletin, Qi$9007). Bailey DG, Mercer KA, Plaw C, Ball R, Barraclough H (2004) High speed weight estimation by image analysis. In Mukhopatlhyay SC, Browne RF, Gupta GS (Eds.), Proceedings of the 2004 New Zealand National Conference on Non Destructive Testing. Palmerston North, New Zealand, Jul 27-29, 89-96. Ballard DH, Brown CM (1982) Image formation. Computer vision, Prentice Hall, Inc. Englewood Cliffs, Ne w Jersey 0763 2, chapter 4, p18 Baltazar A, Espina-Lureco J, Ramos-Torres I, Gonzalez-Aguilar 0 (2007) Effecot of methy jasmonate on properties of intact tomato fruit, monitored with destructive and non-destructive cuts. J Food Engi $b: 1086-1095. Bartlett SL, BesI PJ, Cole CIS, Jain R. Mukherjee D, Sldfstad D (1988) Automatic solder joint inspection. Trans IEEE Pattern Anal Mach Intel! 1(l0): 31-43 Batchelor MM, Searcy SW (1989) Computer vision determination of stem-root joint on processing carrots. 3 Agric Eng Res 43:259-269. Bato PM, Nagata M, Cao QX. Hiyoshi K. Kitahara T (2000) Study on sorting system for strawberry using machine vision (part 2): development of sorting system with direction and judgement functions for strawberry (chime variety). J Jpn Soc Agric Machinery 62(2): 101-110 Saxes GA (1994) Digital image processing principle and applications, New York: John Wiley & Sons. Bender RJ, Brecht JIB, Sargent SA, Huber DJ (2000) Low temperature controlled atmosphere storage for tree-ripe mangoes ( angifera indica L), Arta Hont 509:447-458. Bennedsen BS, Peterson DL, Tabb A (2005) Identifying defects in images of rotating apples. Computers and Electronics in Agriculture 48:102. Bern CJ, Charity LF (1975) Airflow resistance characteristics of corn as nfluereel by bulk density. ASAIB, St. Joseph, ML Paper Igo, 75-3510. Bhanu B, Lee S, Ding J (1995) Adaptive image segmentation using a genetic algorithm. IEEE Transactions on Systems Man and Cybernetics 25 (12):1543-1567. Bharti MH, Liu JJ, MacGregor IF (2004) Image texture analysis: methods and comparisions. Chemornetr Intell Lab Syst 72:57-71. Bland JIM. Altman D (1986) Statistical methods for assessing agreement between two methods of clinical measurement. Lancet., 1:M)7-10. Bland JM., Altman DG (1999) Mca uring agreement in method comparison studies. Stat. Methods Med. Res. 8, 135-160, Bland TM, Altman DG (2007) Agreement between methods of measurement with multiple observations per individual. Journal of Biopharmaceutical Statistics. 17:571-582. Blasco J, Cubero S. Gomez- ancbis J, lira PMolto E (2009) Development of a machine for the automatic sorting of pomegranate (Punica granatam) arils based on computer vision. J Food Eng 90:27-34. flolhouse V (1997) `.Fundamentals of Machine Vision", aobotic Industries Assn. Proceedings of Coastal Sediments' 7. ASCE, New York p 469-483. Bose TI., Miira 8K (2001) Fruits: Tropical and subtropical. Vol 1, Naya Udyog Calcutta, India. Boukbarouba 5, Rebordao TM,'Wendel PL (1985) An amplitude segmentation method based on the distribution function of an image. Comput. Vision Graphics Image Process. 29: 47-59. Brodrick UT (1978) Methods for control of anth coo and other diseases of mango. In: Zehr El a a1., ed. Methods for Eva1uitMg Plant Fungicides, Nematicides, and Bacierkides. St Paul, MN, USA. APS Preys, 80-1. Brosnan T, Sun D-W (2004) Improving quality inspection of food products by computer vision—a review. Journal of Food Engineering 61: 3-16. Bulanon DM, Kataoka T, Ota Y, Hiroma T (2001) A Machine Vision System for the Apple Harvesting Robot. Agricultural ngineeriag Internatiorial: the CKiR Journal of Scientific Research and Development. Manuscript PM 01 006 Vol. III CAC (2004) Recommended international code of practice for packaging and transport of tropical fresh fruits and vegetables. Codex Alinientarius Commission. CAC ! RCP 44 –1995, AND 1-2004 Canieto AFL (2004) Manual for the preparation and sale of fruits and vegetables From field to market. Food and Agriculture Organization of the United Nations service buIletin 151, ch 5, p 91. 243 Cangi R. Altuntas E. Kaya C, Saracoglu (2011) Some chemical and physical properties at physiological maturity and ripening period of kiwifruit ('Hayward'). African J Biotechno110(7): 5304-5310. Canny J (1986) A computational approach to edge detection_ IEEE Trans Pattern Anal Mach IntelIigency ():679 98. doi: 10.1109/TPAMI,198 .47 7851. Cao L, Shi ZK, Clieng EKW (2002) Fast automatic multilevel thresholding method. Electron. Lett. 38: 868-870. Capone S, Distante C, Fraucioso L, Preside D, Taurino AM, Siciliano P, Zuppa (2005) The electronic nose applied to food analysis. An. Moe. Quiui.. Argent. 93: 1-3 Casady W , Paulsen MR, Reid JF, Sinclair JB (1992) A trainable algorithm for inspection of soybean seed quality. Transactions of the ASAE 35:2027-2034. Casasent D, Talukder A, Keagy P. Sehatzki T (2001) Detection and segmentation of items in X-ray imagery. Transactions of the ASAE 44:337-345. Celik A, Ercisli S (2009) Some physical properties of pomegranate cv. Eksinar. Int. Agrophysics 23: 295-298. Cernito E, Faille S. Schiliaci 0 (1996) Identifrcatio of blemishes on oranges. International conference on agricultural engineering, AgEng'96, Madrid, Spain, Eur Agric Eng Paper no. 96 F-017, Sept, 2 . Chan LM, Tan R, Thin (3 (2007) Design of visual-based color classification system JA SA .:30-33. Chao K, Park 13, Chan YR, Hru.schka WR. Wheaton FW (2000) Design of a dual-camera system for poultry carcasses inspection. App2 Eng Agric 16(5):581-587, Chen P. Liang T, Yau H, Sun W, Wang N. Lie H, Lien R (1998) Classifying textile faults with a back propagation neural network using power spectra, Weill/e Research Journal, 68:121-126. Chen, -R., Chao, K., Kim, MS (2002) Machine vision technology for agricultural applications. Computers and 1lectronics in Agriculture, 36: 173-191 Chien Po-J, Shau F, Yang F-H (2007) Effects of edible chitosan coating on quality and shelf life of sliced mango fruit. J Food Engi 78(1): 225-229 Choi K, Lee , Han YJ, Bunn J (1995) Tomato maturity evaluation using colour image analysis. Transactions of the A SAE 38:171-176. Chtioui Y, Panigrahi S. Backer LF (2003) Self-organizing map combined with a fuzzy clustering for color image segmentation. Transactions of the ASAE 46 (3);831- 838. ober ER , Voldeng HI) Fregeau-Reid JA (1997) Heritability of seed shape and seed size in soybean. Crop Science 37: 1767-1769. 244 Corlddi G, Balderas-Ruiz KA, Tahaada B, Serrano-Carreoa L, Galindo E (2006) Assessing mango anthracnose using a now three-dimensional image-analysis technique to quantify lesions or frixiL Plant Pathology 55,250-25.7 Cortez F, Portelinha M (2O0) Lamb meat quality assessment by support vector machines, neural processing letters, Springer, Inpress, 1S N. 1370-,4621 Crowe TG, Deiwiche MJ (1996a) Real-time defect detection in fruit-Part I: design concepts and development of prototype hardware. Transactions of the ASAE, 39 (6): 2299-2308. Crowe TG,. De1wiche ICJ (1 996b) Real-time defect detection in fruit - Part U: an algorithm and performance of a prototype system. Transactions of the A AE, 39 6): 2309-2317. Daley W, Carey R, Thomson C (1994) Real-time coIour grading and defect detection of food products. Optics in agriculture, forestry and biological processing SPIE (International Society for Optical Engineering). Int Soo Opt Eng 2345: 403-411_ Daley WD, Thompson J (1992) Color machine vision for meat inspection, Sri; Food processing automation I1—proceedings of the FPAC conference, ASAB. St. Joseph. Michigan, p. 230-237. Danno AM, Miyazato, Ishiguro E (1980) Quality evaluation of agricultural products by infrared imaging method: Maturity evaluation of fruits and vegetables. Memoirs of the Faculty of Agriculture, Kagoshima University 16: 157-164. Das K, Evans MI (1992) Detecting fertility+ of hatching eggs using machine vision. H. Neural network classifiers. Trans ASAE 35 () 7035-204X. Davenel A, Guizard C, Labarre T, Sevila F (1988) Automatic detection of surface defects on fruit using a vision system. J Agric Ing Res 41: 1-9. Davidson VJ, Ryks 1, Chu T (2001) Fuzzy models to predict consumer ratings for biscuits based on digital features. IEEE Trans Ftizzy Syst 9(1): 62-67. Davies ER (1997) Machine vision, 2t d edn, Academic, New York. Davies ER, Bateman M, Mason DR, Chambers J (2003) Pattern Recognition Letters 24(1-3):41 -428. Davis AP Lettington AH (1988) Principles of thermal imaging, In Applications of Thermal Imaging ed. SG Burney, TL Williams, C,H, Jones, 1-34. Bristol: IOP Publishing Ltd, Do Oliveira Costa VS. Mich ref SJ, Martins RB, (lava CAT. Mizubuti E G, Camara MPS (2010) Species of Botryosph eriaccae associated on mango in Brazil. Eur. 1. Plant Path l, 509-519, DeE11 JR, Vigncault C, Lemerre S (2000) Water temperature for hydro-cooling field cucumbers in relation to chilling injury during storage, Postharvest Rol. and Tech. 1827-32. 245 Delwiche MJ (1987) Grader performance using a peach ground calm maturity chart J Americ Soc Hort Sci 22(1): 87-89. Delwiche 13J, Baumgardncr R A (1983) Ground color m ►surement of peach, I Am Soc HortieSci 108: 1012-1016. Dewulf W, Jancsuk P, Nicolai B, De Roeck CO, Briassoulis D (1999) Determining the firmness of a pear using finite element model analysis. J Agric Eng Res 74(3): 217-224. Diaz R. Fans G, Blasco M, Blasco J, Motto E (2000) The application of a fast algorithm for the classification of olives-by machine vision. Food lies Int 33: 305-309. ' Diedhiou PM, Mbaye N, Draxnel A, Samb P1(2007) Alteration of post harvest diseases of mango Mang ferci indica through production practices and climatic factors. African Journal of Biotechnolo y, 6 (9): 1087-1094, Ding CK, Chacliin K, Ueda Y, Imahori Y, Wang CY (2002) Modified atmosphere packaging rnaintaiins postharvest quality of loquat fruit. Postharvest Biel Technnl 24(3): 341-348 Dodd JC, Estrada AB, Matcham J, Jeffries P, Jeger MJ (1991) The effect of climatic factors on Colletotrichum gloeosporioidcs, causal agent of mango anthracnose, in the Philippines, Plant Pathology 40, 8-75. Dorvlo SY, Bani RJ, Sinayobe E (2012) Prediction of volume, weight and surface area of banana (M.isa uniriata) using picture image analysis, Journal off Food, Agriculture & Environment Vol.10 (3&4) 112-114. Dowell FE, Throne JR, Wang 11), Baker JE (1999) Identifying stored-grain insects using near-infrared spectroscopy. J of Economic Entomology 92: 165-169. Du J, Sun D-WW (2004x) Recent developments in the applications of image processing techniques for food quality evaluation. Trends in Food Science & Technology 15:230-249. Du J, Sure D- (2004b) Shape extraction and classification of pizza base using computer vision, Journal of Food Engineering, 64 (4), 489-4%. Du -J, Sun D- (2006a) Automatic measurement of pores and porosity in pork hang and their correlations with processing tune, water content and texture. Meat Science 72(2) 294-302. Du -J, Sun D-W (2006b) Learning techniques used in computer vision for food quality evaluation- a review. Journal of Food Engineering 72:39-55. Du C-J, Sun D- W, Jackman P. Allen P (2008) Development of a hybrid image processing algorithm for automatic evaluation of intramuscular fat content in beef m. loo issimus dorsL Meat Science 80: 1231-1237, 246 Dugo P, Mondello L, Favoino 0, Cicero L, Rodriguez enteno NA, Dugo G (2005) Characterization of cold-pressed Mexican Dancy tangerine oils. Flavour Fragr. J. 2G: 60--6& Ehrahimi E. Mollazadn K, Arefi A (2011) Detection of greening of potatoes using image processing techri ques. Journal of American Science 7(3) 243-247. Eifert JD, Sanglay GC. Lee DJ, Sumner SS,- Pierson MD (2006) Prediction of raw produce surface area from weight measurement X of Food Engg 74:552-556. FatholahzaOeh H, Rajabipour A (2008) Some mechanical properties of barberry. int. Agrophysics :299-302.. - Faucitano I., Huff P, Tcuscher F, Cariepy C, Wegner J (2005) Application of computer image analysis to measure pork marbling characteristivs. Meat Science 69(3):537- 543. Fava J, 1-lodara K, Nieto A, uerrero S. Aizamora SM1 Castro MA. 2011. Structure (micro, ultra, nano), color and mechanical properties of Vitis tabrusca L. (grape berry) fruits treated by h}rd.rogeu pmxide, UV irradiation and ultrasound. Food Res ant, In Press. Gaffney 33 (1973) Reflectance properties of citms fruit. Trans ASAE 1-6(2):310-314. Sauch JM (1992) Investigations of image contrast space defined by variations on histogram equalization. VGIP: Graphical Models of Image Process 54(4): 269- 2g0. Gerrard OF, Ciao X, Tau J (199) Beef marbling and colour score determination by image processing. J of Food Sc (1): 145-148. Glizanfari A, Trudayaraj J, Kusalik A (1996) Grading pistachio nuts using a neural network approach. Trans ASAE 39 (6): 23 I9-2324 Gill PS, SingFi P , Jawandha SK (2005) Post harvest handling of mango- a review. Proceedings: Intt conference on mango and date palm: culture and export, loth to 23cdJi.me. Milk et al. (eds), University of Faisalabad. 001 (0I3) Post-harvest profile of mango, ministry of agriculture (department of agriculture &: cooperation) Directorate of Marketing & Inspection branch head office Nagpur, I-141, Gomez A. Wang J, Perim AG (2005) Impulse response of pear fruit and its relation to Magness-layler firmness during storage. Post Harvest Biology and Technology 35: 209-215. Gonzales RC, Woods RE, Eddins SL (2004) Digital image processing using MATLAB. Prentice-Hall. Upper Saddle River, New Jersey. Gonzalez RC, Woods RE (2002) Digital image processing. Introduction, chi, 2nd edition, Ch. 1. Prentice-HaII, Inc, Upper Saddle River, Now Jersey. 247 Goodrum YW, Sister RT (1992) Machine vision for crack detection in rotating eggs. Transactions of the ASAE 35: 132x132 4. Gowda IND, Huddar, AG (2001) Studies on ripening changes in mango ( angf fera indica L) fruits. J Food Technol 38 (2): 135-137. Gowda IND, Ramanjancya Ki-1 (1994) Studies on ph s co-chemical characteristics of some commercial cultivars of mango, Indian Food Packer 50(2), 45-49. Goyal RK, Kingsly ARP, Kumar P, Walia H (2007) Physical and mechanical properties of aonla fruits. r Food Engi 82: 595-599. Guillermin F; Dupont N, .e Morvan C, Le QueTe JM, Langlias C, Mauget JC 2006) Rheological and technological properties of two cidar apple cultivars. LWT, 39: 995-1000. Grurnus B, Balaban MO, Unlusaym M (2011) Machine Vision Applications to Aquatic Foods: A Review. Turkish Journal of Fisheries and Aquatic Sciences 11: 171- 181. Gunasekaran S (1996). Computer vision technology for food quality assurance. Trends in Food Science and Technology, 7(8): 245-256. unaselcaran S (2001) Non-destructive food evaluation techniques to analyse properties and quality, food science and technology series. Marcel Dekker, New York, p 105. Gunasekaran (2008a) Quality evaluation of cheese. Computer vision technology for food quality evaluation-edited by Da-Wen Sun, Director, Food Refrigeration and Computerised Food Technology Rrese=h Group, National University of Ireland, Dublin, Ireland. ISBN: 978-0-12-3736424), p457. Gunasekaran S (2008b) Quality evaLuation of -cheese. Coputer vision technology-edited by Da-Wen Sun p457 Gunasekaran S. Ding IC (1993) Using computer vision for food quality evaluation. Food Tech of 6 :151-154. Baciscferogullari H, Gamer I, Ozcan MM Murat srn B (2007), Post harvest chemical and physical-mechanical properties of some apricot varieties cultivated in Turkey. J Food Engi 79: 364-373. Haciseferogullari H, Ozean M, So mete M I, Ozbek 0 (2005) Some physical and chemical parameters of wild rrjedlar ( M.spilus gen r ica L.) fruit grow in Terkey, Journal. of Food Engineering 69. 1-7. Hagen SF, Soihang KA, Bengtsson GB, Gorge CIA, Bilger 1 (2006) Chlorophyll fluorescence as a tool for non-destructive estimation of anthocyanins and total flavonoids in apples. Posthary Biel Technol 41: 156-163, Hahn F (2002 Automatic Jalapetio Chilli Grading by Width", Biosysteem Engineering, 83(4), pp. 433--440. 248 Hahn F, Sanchez S (2000) Carrot volume evaluation using imaging algorithms. J Agric Eng Res 75:243-249. Haralick RM (1979) Statistical and skructural approaches to texture, Proceeding of the IEEE 67:786--504. Hardenburg RE, Watada AE, \ ang CY (1986) The Commercial Storage of Fruits, Vegetables, and Florist and Nursery Crops. In: USDA Agriculture Handbook No. 66 72. Beltsville. Harding PL, Soule MJ, Sunday MB (1954) Quality in mangos. Prot. F'orida Mango Forum, pp. 28-33. Harrabi R, Ben Braiek E (2011) Color image Segmentation using automatic thresholding technique, 8th International Multi-Conference on Systems, Signals &. Devices, 978-1-4577-0411-6/11, LEEE1 - Harrell k (1991) Processing of color images with bayesian discriminate analysis. 1st intent ional seminar on use of on-machine vision systems for the agricultural and bio-industries, CEVLkGREF Montpellier, France, 3-6 Sptembcr, pp 11-20. Hashim MS, Ayub M, Zeb A, Hashim MM. Alain 8, Riaz A (2006) Effect of chemical preservatives on the physicorche.mical and nuritional evaluation of mango pulp stored at ambient temperature. Sarhd J Agric 22(4): 685-681. Hashim N, Janius R, Baranyai M, Pflanz M, Regen C, Zude M (2010) Application of RGB and Backscuttering Imaging to Detest Chilling Injury Symptoms in Banana. CIGR Workshop on image Analysis in Agriculture. 25-2']. August 2010, Budapest Hatcher DW Kruger JH, Anderson MJ (1999) Influence of water absorption on the processing and quality of oriental noodles. Cereal Chemistry 76, 566-572. Hatcher DW, Symons SJ (2000) Image analysis of Asian noodle appearance: impact of hexap laid wheat with a red seed coat. Cereal Cheinisfry 77, 388 391. Elatern I, Tan 1, Gerrard DE (2003). Determination of animal skeletal maturity by image processing. Meat Sci 65:999-1004. He Y, Shang Y. Pereira AG, omez All, Wang .1(2005) Noidetructive 1 ter ination of Tomato Fruit Quality Characteristics Using is/I IR Speciroscapy Technique. Interuat oral Journa[ of Information Technology l 1(11): 97-148. Heinemann PH, Hughes R, Morrow T, Summer HJ, Beelman RB, NVuest PS (1994) Grading of mushrooms using a machine vision system. Transactions of the ASAE 37 (5) 1671-1677. Heinemann PH, Varghese ZA, Morrow CT, Sommer HJ III, CrasswIler RM (1995) Machine vision inspection of "Golden De'icious" apples. Appi Eng Agric 11(6): 901-906. 249 Hellehrand HJ, Douche ML, Herold B and Geyer M (2001) Chances and Shortcomings of Thermal Imaging in the Evaluation of Horticultural Products. Institute of Agricultural Engineering Bornim (AT13), Max-Eye-A.11ee 100, 14469 Potsdam, Germany. Howarth MS, Searcy SW (1992) Inspection of frcsb carrots by machine vision. In: Food Processing Automation II Proceedings of the 1992 Conference, ASAE, 2950 Niles load, St. Joseph, MI 49085-9659, USA. Luang Q, Gao W, Cai W (2005) Thresholding technique with adaptive vwindow'selection for uneven lighting imago. Pattern Recognition Letters 26: 8018808. Hulbert GJ, Litchfield JB, SJ Schmidt (1995) Temperature mapping in carrot using Ti weighted magnetic resonance imaging. J load Sci 60: 760-785. Huhne AC (1971). The mango. In: Biochemistry of fruits. and their prciducts. Academic Press, Loudon and New York, pp 233-253. Hwang H, Park B. Nguyen M, Chen -R (1997) Hybrid image processing for robust extraction of lean tissue on beef cut surface. Comput Electron Agric 17(3):281- 94. Iagtiani J. Chan lid', William J (1988) Ttopieal fruit processing. Academic press INC. Harcourt Brace Jovanovich, Publishers San Diego, New York, Berkly, Boston, Landon, Sydney, Tokyo, Toranto, pp. 52-73 Igual M. Castello ML1 Ortola MD, Andres A (2008) Influence of vacuum impregnation on respiration rate, mechanical and optical properties of cut persimmon. J Food Engi 86(3): 315-323 Iwamoto M, Kawano S, Yukihiro 0 (1995) An overview of research and development of near infrared spectroscopy in Japan. J. Near Infrared Spectrosc. 3: 179-189. Jackman P, sun D-W, Allen P (2011) Recent advances in the use ofr computer vision technology in the quality assessment of fresh meats. Trends in Food Science & Technology, 22(4), 185-197 Jacobi K , Wonga LS, Giles JE(1995) Effect of fruit maturity an quality and physiology of high-humidity hot air-treated Kensington' mango (Mangijkwi Indira Linn.). Postharvest Biol Technol 5: 149-159 Jahns G, Nielsen HM! Paul W (2001) Measuring image analysis attributes and modelling fuzzy consumer aspects for tomato quality grading. Computers and Electxorjics in Agriculture 31:V7-29 Jain AK (1989) Fundamentals of digital image processing. Englewood Cliffs: Prentice- Hall. Jain NC, Paape MJ, Mille- RH (1991) Use of flow cytometry for determination of differential Ieulcocytc counts in bovine blood. Ann J Vet Res 52(4): 630-636. Janick J (1986). Horticultural Science, 4th edn. New York Freeman, WH, 550-5 1. 250 Jarimapas B, Jaisin N (2008) An experimental machine vision system for sorting sweet tamarind.. Journal of Food Engineering 89: 291.297, Jha SN, Chopra 5, Kingsly ARP (2005) Determination of sweehiess of intact mango using visual spcetral analysis. Biosystems Engg 91: 157-161. Jha SN, Kingsly ARP, Chopra S (2006) Physical and mechanical properties of mango during growth and storage for determination of maturity. J Food l ngi 72: 73-76. Jha SN, Matsuoka T (2060) Non-Destructive Techniques for Qa1ity Evaluation of Intact Fruits and Vegetables, Food Sci Technol lies 6 (4): 24&-251. Jha SIN Matsuka T (20041 Nonde trucpirre determination of acid brLx ratio of tomato juice using near infrared spectroscopy. Inter~atiomal 3 Food Sci Technol 39: 425- 430. Tha SN, Narsaiah K, Sharma AD, Singh M, Bansal 8, Kumar R (2010), Quality parameters of nn.ango and potential of non-destructive techniques for their me urement - a review. I Food Sci Technol 47(1): 1-14. Ji ZL., Znang ZQ, Wang Y, Li XP (1994) Low temperature storage and chiIing injury of mangoes- ActaHort inica 21:l 11-11 . Jia P. Evans NOD, (Tate SR (1996), Catfish feature identification via computer visions. Transactions of the ASAE 39:1923-1931. Siang, X., Mojon, D (2003) Adaptive local threslolding by verification-bid multithrisliold probing with application to vessel detection in retinal inuges. IEEE Trans. Pattern Anal, Machine bitel1. 25,131-137. Sianjun Y, Ilanping M, Yongguang 3 Xiiizong W, Shuren C (2006) An Automatic Segmentation Method for Multi-tomatoes Image under Complicated Natural Background, Proc. of SHE Vol. 6411 641118-1, doi: 10.1117112.697799. Ting G. kajan D, Siang CE (2005) Motion detection with adaptive background and dynamic threshold. In IEEE bternt Cortf Information, Communications and Signal Promising, Bangkok, Thialand, December, pp.AI-45, Jirsa 0, Hruskova M, Svec 1(2007) Bread features evaluation by NM analysis_ Czech I Food Sci.25:243-248. John J. Subbarayan C, Canna HR (1970) Carotenoids in three stages of ripening in mango. J Food Sel., 35: 262-265. Johnson GI, Cooke ASV (1991) Stem end rot mango in Australia: cause and control, I H5 Acta. Reif. 291: 288-295. Johnson PR, Parr I) (20-46) Mango growing in Western Australia. Department of agriculture and Food, Western Australia, Bulletin 4348, Jungalwala FB, Carna QTR (1963) Carotenoids in mango ( g(era indica) fruit. Indian Journal of Chernishy 1, 36-40. 251 Kabas 0,- Ozmerzi A, Akinci 1(2006) Physical properties of cams pear (Opuntia ficus India L.) grow wild in Tummy. J food Engi 73: 198-202 leader AA, Sommer NF, Arpala ML (2002) Postharvest handling systems: tropical fruits. In: A.A.Kader (ed), Postharvestvest technology of hoRicultural crops, third edition.University of California, Agriculture and Natural Resources, Publication 3311, pp.385-398 Kanali C, Murase H, Honami N (1998) 'Three-dimensional shape recognition using a chargesirnulation method to process image features. I Agri Engg Res 70: 195- 208. Kane 0, Macellm P (1978) Incidence of ripening and chilling injury on the oxidative 'activities and fatly acid composition of mitachondriya from mango fruits. Plant ph 5i ul, 61:634-640. Kantt CA, Schmidt SJ, Sizer SE. Palaniappan S. 1B Litchfield (1998) Temperat= mapping of particles during aseptic processing with magnetic resonance imaging. Journal of Food Science 63(2): 305-31-1 Kautt CA, Webb AG, Litchfield JB (1997) Temperature measurement of foods using chemical shift magnetic resonance imaging as compared with T1- weighted temperature mapping. J Food Sci 62(5): 1011-101 . Kapur JN, Sabo PIS, Wong AK (1985) A new method for grey level picture thresholding using the entropy of the histogrant Compost Vision Graphics and Image process 29:273-285. Kato K (1997) Electrical density sorting and estimation of soluble solids content of watermelon, lAgricEng lies 67: 161-170. Katsumata T, Aizawa H, Matashige E (2004) In the Extended Abstract of 206th meeting of the Electrochemical Society, Honolulu Hawaii oct. Katsumata T. Aizawa H, Saito M, Komuro S. Morikawa T I 1~ ICA E (2006) Int Joint Conference issue 18-21: 1578 —1581. Kaur N, Kaur P (2011) A review on various methods of image thresholding. International Journal oii Computer Science and Engineering (EJ E), 3(10) 3441-3443. Ketsa 55. Chidira , Klein JD, Lurie S (1999) Ethylene synthesis in mango fruit following heat treatment. J Postharvest Biol Technol 15: 72, Khanali MM, Ohasemi Varnamkasti A. Tabatabaecfar,Mohli H (2007) Mass and volume modelling of tangerine (Citrus reticulate) fruit with some physical attributes. Int. Agrophys. 21:32 334. Khezri L, Rashidi M, Gholami M (2012) Modeling of Peach Mass Based on Geometrical Attributes Using Linear Regression Modeis, American-Eurasian J. Agric. & Environ. Sci., 12 (7): 991-995. I iojastelnazharxd M, Amid M. Tabatabaee rA (2009) Determination of orange volume and surface area using image processing technique. Int Agrophys 23: 237-242. 22 hoja tehnazhand M, Omid M, Tabatabaeefar A (2010) Development of a lemon sorting system based on color and size. Afr J Plant S ci 4(4): 122-121. Khoshnam F, Tabatabaeefar A, Gh erni-Vamamkhasti M, Burgher AM (2007) Mass modeling of pomegranate (Punica granatum L) fruit with some physical chaicthristks. Sci. Hortic. 114: 1 6. ilic K. Boyaci [H. Koksel H, Kusmenoglu 1 (2007) A classification system for beans using computer vision system and artificial neural networks, Journal of Food Engineering 78: 897-904. K.ilinoja L (0] 0) Identification of appropriate postharvest technologies fDr Improving - market access and incomes for small horticultural farmers in Sulu-Saharan Africa and South Asia. Final Report, Alexandria VA: WFLO ittler J, ]'lhingworth J (1 9S6) Minimum error threshoI.ding, Pattern Recognition 19(1): 41-47. K2eynen 0, Leemans V, Destain M-F. 2005. Development of multi-spectral vision system for the detection of defects on apple. Journal of Poo d Engineering 69: 41- 49. Knight TO, Klicbt A, Sedgley M (2002) Structural basis of the rind disorder -oleocellosis in Washington navel orange (Citrus sinensis L. Osbeck). Ann. Bot. 90: 765--773. Ko Y, Liu CW, Chen CY, Marathasalam S, Lin CE (2009) First report of stem-end rot of mango caused by Phomopsis wrong;/erue in Taiwan. Plant Dis. 93: 764. Kubil€:r 1, 1 ved R. Artez L, Prusky D, Johnson Gil, Highly 1E, Joyce DC (1995) Antifungal compounds regulating quiescent disease in mango. Disease resistance in fruit. Proceedigs Of An international Workshop Held At Chiang Mia, Thailand 1997. ACL4R Proceedings Series 80:109-114, I obiler I, Shalom Y, Roth 1, Akerrnan M. Vinokni Y, Fuchs Y, Presk U (2001) Effect of 2,4ich1oropbeoxy acetic acid cn the incidence of side and stem end rots in mango fruits. Postharvest Biology and Technology 23: 23-32. Coo AB (2007) Determination of Watermelon Volume Using EIlipsoid Approximation and Image Promising. I Postharvest Bill Technol 45: 366-371, Kondo N (1995) Quality evaluation of orange fruit using neural networks. In: Food Processing Automation IV Proceedings of the FPAC Conference. #SAE, 2950 Niles Road, St. Joseph, MI 49085-9659, USA Kondo N, Ninomiya K. Peter R1 Kamata J, FasiI A (200.) Development of Multi-Product Grading System, A AE Annual Meeting, paper No. 043125. Koomen I, Jeffries P (1993) Effects of antagDnistic microorganisms on the post harvest development of Collelotrichum g!oeospori*ides on mango. Plant Fathoiogy 42: 230-7. 2 3 Kosehan A, Abidi M 2001) A Comparison of Median Filter Techniques For Noise Removal in C 1nr Images. Proc. 7th Gem1an Workshop on Color Image Processing, D. Paulus, .J. Henzler (Eds.), 'Erlangen, Germany, Institute of Computer Science, October, 34(15): 69-79. Kritz GW, Gibson H , Cassens DL,, Zhang M (2000) Colour vision in forest and weed engineering. landwards 55(1):2--9 Warren DE (1997) Image analysis research at NIB (National Institute of Agricultural Botany): chrysanthemum leaf shape. Plant Var Seeds 10(1): 59-61. Kudachikar VB. Kuikami SG, Kcshav PMN, Vaaantha I S,B Parsad SA, Ramaii KVR - - (001) Physi --chemical changesduring maturity of mango ( angi era indicia Q var. Ncelum'. J Food Sci Technol 38: 540-542 Labbafi M, Thakur RIB, Vial C, Djelveh C (2007) Development of an ors-line optical method for assessment of the bubble size and morphology in aerated food products. Food Chemistry 102: 4544-65. Lak MB, Miinaei 5, Amiriparian J, Beheshti B (2010) Apple Fruits Recognition Under Natural Luminance Using Machine Vision. Advance Journal of Food Science and Technology 2(): 325-327 Laid HiD, Singh Z, Tan SC (2001) Elevated level of CO2 in controlled atmosphere storage affects the shelf life, carotenoid and fruit quality -of mango. Acra Hoyt. 553:605-606 Lalel HiD, Singh 2, Tan SC (2003) Elevated levels of CO2 in controlled atmosphere ,storage affects shelf life, frrxit quality and aroma volatiles of mango. Acra Hart. 628:407413 Law hLJD, Singh Z, Tau SC (2005) Controlled atmosphere storage af#'ects fruit ripening quality ofbeelta R2E2' mango. J Hurt. Sc. Biotechnol 80: 551-556. Lam L, Lee SW, Suers CY (1992.) Thinning methodologies-a comprehensive survey. IEEE Trans Fattarn Anal Mach Intel 14 (9):869-8S5 Laurent 13, Ousman B, Dzudie T. Carl I+ fM, Emmanuel T (2010) Digital camera images processing of hard-to-cook beans. J Engg Tech Res 1012010; 2:177-188. Leemans V. Destain MF (2004) A real-time grading method of apple based on features extracted from defects. Journal of Food Engineering 61:83—S9. Lc-emans V, Magein H, Destain MF (1998) Defects segmentation of `"Golden Delicious" apples by using colour machine vision. Computers and Electronics in Agriculture 20 (2):l17-130. Leemans V. Magein H. Destain MF (1999) Defects segmentation on "Golden Delicious" apples by using colour machine vision and Bayesian classification method. Computers and Electronics in Agriculture 23 (1): 43-53. 254 Lefebvre M, (311 DS, Brwiet E, Natonek C, Baur P, Gugerli, Pun T (1993) Computer vision and agricultural robotics for sease control: The potato operation, Comput F1ectron Agric 9: 85l02. LeOn K, Mery I), Pedreschi F, Leon J (2006) Color measurement in L*a*b* units from }KGB digital images. Food Research International 39: 1084-1091 Li IF Tan J, Martz FA 19 7a) Predicting beef tenderness from image texture features. Fn 1997 ASAB Annual International Meeting, Paper No. 973124. St. Joseph, Michigan, LISA: ASAE. Li J, Tan.J, Martz FA (199Th) Predicting beef tenderness from image texture features. In: - 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973124, ASAE, 2950 Niles Road, St. Joseph, Ml 49085-9659, USA. Li J, Tan J, Martz FA, Heymann H (1999) Image texture features as indicators of beef tenderness. Meat Science, 53, 17-22. Li 3, Tan J, Shatadal P (2001) Classification of tough and tender beef by image te.xture analysis. Meat Science 57:341-346. - Li QZ, Wang MH (1999) Study on high-speed apple surface defect segment algorithm based on computer vision. In: Proceedings of International Conference on Agricultural Engineering (99-ICAE), Beijing, People's Republic of China, 14-17 December, p \'27-31. Li QZ, Wang Mil, Gu WI (2002) Computer vision based system for apple surface defect detection. Computers and Electronics in Agric iltur'e 36(2-3):215-223. Liao, J (1975) Mango disease in Taiwan-stem-end rot. Scientific Agriculture 23: 415- 416. Lichtenthaler HK, Lang M, Sowinska M, Heisel F. Miehe IA (1996) Detection of vegetation stress via a new high resolution fluorescence imaging system. J Plant l~hysi 148: 599-612. Liu L-f, Lu F-M, Chang YC (2006) Development of a Near-Infrared Imaging System for Determination of Rice Moisture. Cereal Chemistry 83 (5): 498-504. Ling FP, Ruzhitky VN (1996) Machine vision techniques for measuring the canopy of tomato seedling. J Agric Eiag Res 65: -995. Lino A L, Sanchcs J. Fabbro IMD (UD$) Image processing techniques for lemons and tomatoes classification. Bragantia campinas 67(3): 785-789. Liu J, Paulsen MR. (1297) Cow whiteness measurement and classification using machine vision. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973045, ASAE, 2950 Niles Rand, St. Joseph, MI 49085-9659, USA, Liu \V, Tao Y, Sieberimorgon TJ, Chen H (1997) Digital image analysis method for rapid measurement of rice degree of milling. In: 1997 ASAE Annual International 255 Meeting Technical Papers, Paper No. 973028, ASAE, 2950 Niles Road, St Joseph, Ml 49085-9659. USA. Lloyd BJ, Cnossen AG, Siebeninorgen TJ (2000) Evaluation of two methods for separating head rice from brokens for head rice yield determination. In 2000 ASAE Artnu.al International Meeting, Paper No, 006074 ASAE, 2950 Niles load, St. Joseph, MI 49085-9659, USA. Locht P, Thomsen K, Mikkelsen P (1997) Full color image analysis as a tool for quality control and process development in the food industry. Paper No. 973006, ASAE. 2950 Wiles Road, St. Joseph, Ml 49085-9659 USA, Lyon VPCC (1996) Bet bepalen van het ontwikkelingsstadiuin bij dechampignoai met . computer becldanalyse. Charnpignorxcultuur 40(9): 347-353. Lorestani AN, Kazerni A (2012) Mass Modeling of Castor Seed (Rieinus cornmunis) with Some Geometrical Attributes. International Journal of Agriculture and Forestry 2012. 2(5): 235-238. Lorestani AN, Omit M. Shoorad S13, Borghei AM, Tabatabacefar 4 (2006) Design and evaluation of a fuzzy logic-based decision support system for grading of Golden Delicious apples. Int J Agric B iol 8(4): 440-44.4. Lorestani AN, Omid M. Shooraki SB. Borghei AM, Tabatabaeefar A (2006) Design and evaluation of a Fuzzy Logic based decision support system for grading of Golden Delicious apples. Tflt I Agric Bial 8(4l: 440-444 Lorestani AN, TAbatabaeefar A (2006) Modelling the mass of 1dwifruit by geometrical attributes. Int. Agrophysics 20: 135-139. Loveys BR, robinsonSi', Brophy IJ, chacko EK (1992) Mange sapbum: Components of fruit sap and their role in causing skin damage. Australian Journal of Plant Physiology, 1:44 9-457. Lu J, Ian 1, Gerrard DE (1997) Pork quaIity evaluation by image processing. In: 1991 ASAE Annual International Meeting Technical Papers, Paper No. 973125, ASAE, 2950 Niles Road, Sn. Joseph, Ml 49085-9659, USA Lu J, Tan J, Shatadal P. Gerrard DE (2000) Evaluation of pork color by using computer vision. Meat Science 56: 57-60 Lu R (2003) Detection bruises on apples using near-infrared hyper spectral imaging. Trans ASAE 46 (2): .3-530 Mac Adam DL (1970) Sources of color science.Cambridge: MIT Press. Maduako JN, Faborode MO (1990) Some physical properties of cocoa pods in relation to primary processing, If e journal of Technology 2: 1-7. Mahon B (2003) The Man Who Changed Everything: The Life of James Clerk Maxwell. John Wiley fir. Sons, Ltd, Chichester, West Sussex, UK. 256 Majumdar S. Sa s DS (2000a) Cssiflcaticn of cereal grains using machine vision: I. Morphology models, Trans ASAE 43 (6): 1669-1675. Majumdar S, Jayas DS (20O b) 'assificati of cereal grains using machine vision: II. Color models. Transactions of the ASE 43 (6): 1677-1680. Majumdar S, Jayas DS 200DcJ I.asaif"ication of cereal grains using machine vision: I. Texture models. Transactions of the ASAE 43(6): 1681-1687. Majumdar 8, Jayas DS (2000d) Classification of cereal grains using machine vision: IV. Morphology, color, and texture models. Transactions of the ASAE 43(6): 1689- 1694. - Majumdar S. Jayas 1)3, Bulley NR (1997) Classification of cereal grains using machine vision, part 1: morphological features. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973101, rLSAE, 2950 Niles Road, St. Joseph. MT 49085-9659, USA Maic:how D (2007) imaging in the Shodwwe Infrared (SW1R) for Industrial Applications, Advanced Imaging magazine, February 2007. SU1 (Sensors LTnlimited, Inc.), part of Goodrich Corporation Manickavasagan A, Jayus DS White NDG, Paliwal J (2005) Applications of thermal imaging in agriculture — A review, Canadian Society for Engineering in Agricultultural, Food and Biological Systems Paper no. 05 02. Mann SS, Singh RN (1976) The cold storage life of Deshari mangoes. Seienria Horlrcul rae 5: 249-254. I4 ai sour FS Abd-El-Aziz SA, Helal GA (2006) Effect of fruit heat treatment in three mango varieties on incidence of postharvest fungal disease. J, P1. Pathol. 88(2): 141-148. I Iagbool 9 Malik AU (2007) Evaluation of some potential chemicals for reducing sap bun] injury in mango. Proceeding: rnterntional Symposium on Prospects of Hnrticalhu 1 Industry in Pakistan, 28t -30" March, Institute of Horticultural Sciences, University of A riiulturc, Faisalbad, I arceau 1;' (2009) Advatce imaging,. Acadc ie Journals and books at Questa Online Library, p294. Marique T, Pennincx S, Khamubi A (2005) Image segmentation and bruise identification an potatoes using a hoiieu's seI#-organizing map. Journal of Food Science 70 7): E41 E417. Marquez U, Anon MC (1986) Influence of reducing sugars and am"hio acids in the colour development of flied potatoes. J Food Sci 51: 157-160. Mal- D, Hildretb E (1980) Theory of edge detection. Proceedings oftthe Royal Society of London. Serius B, Biological Scienccs 207 (1167'): 187-217. 757 Martin D (2007) A practical guide to machine vision lighting, Midwest Sales and Support Manager, Advanced IIlumination. http://www.grafiek.cornlpdfFMachine ision. McCauley JD, Thane BR, Whittaker AD (1994) Fat estimation in beef ultrasound images using text= and adaptive logic network. Transactions of the ASAE, 37: 997- 1002. Mc lemernts DI (1995) Advances in the application of ultrasound in food analysis and processing. Tin s in Food Sciences and Technology . 293-299. McClure WF (14$7) Near:'snfrartd jnstmmentation. In P. Williams & K. Norris (Eds.), Near-infrared technology in the agiicultur.t and food industries, St. Path., MN: American Association of Cereal Chemists, Inc. p89—!05. McDonald T, Char YR (1990) Separating connected muscly tissues in images of beef carcass sib es. Transactions of the ASAE. 33(d): 2059-2065, McNitt-Gray M. Fhiang H , Sayre WI (1995) . Feature selection - in the pattern classification problem of digital chest radiograph segmentation. IEEE Transaction Medical Imaging 14(3): 537-547. Medina J De La C, Garcia H (2002) Mango: Post-harvest Operations, Food and Agricultural Organization of the United Natioi s. Medlicott A, BliogoI M, Reynolds S (1986) Changes in peel pigmentation during ripening of mango fruit (Magnifera indica, oar. ToLmny Atkin), Annals of Applied Biology 109, 651-656. Mdlicott A, BhogoI M, Reynolds S (1996). Changes in peel pigmentation during ripening of mango fruit (Magnifera indica, var. Tommy Atkin). Annals of Applied Biology 109, 651-656. Med.licott ,AP (1990) Ripening of mangoes following low temperature storage. J. Amer. Soc. fort. Sd., 115(3):430-434. leddlicott AP, Semple T, Thornpsori AJ, Black BHEI Thompson AK (1992) Mea.uremept of color changes in nperirg bananas and mangoes by instrumental, chemical and visual assessments. Tropical Agriculture (Trinidad and Tobago) 69: 161-166. Mehl PM, Chen YR, Kim MS, Chan DE (2004) Development of hyper pectxaI imaging technique for the detection of apple surface defects and contaminations. J of Food Eugg 61: 67-81. Mendoza F, Dejtnek P. Aguilera JM (2007) Colour and image text analysis in classification of commercial potato chips. Food Res Tut 40:1146—I 154. enezes JB, Alves RE, Freire FCC) (1995) Mango sapbum-a postharvest injury. R. Brasiiian Fisiol Veg 7: 81-484 Miller B , Delwiche MI (199Ib) Peach defect detection with machine vision. Trans. ASAE 34 2588-2597. 258 Miller MK, Deiwiche MJ (1989) A colour vision system for Peach grading. Trans ASAE 32(4):1484-1490. Miller WM, Peleg K, Briggs P (1958) Automated density separation for fr z d waged citrus. Applied Engineering in Agriculture 4 (4) 344-345.. Miskelly DM (1993) Noodles-a new look at an old food. Food Ault 45: 496-500 itra S1(, Baldwin EA (1997) Mango. In: S. Mitra (ed.). Postharvest physiology and storage of tropical and subtropical fruits. CAB International, Wallingford, UK pp. 8-122, Mohsenin NN (1978) Physical properties of plant and animal materials, Gorden and Breach Science Publisher, New York. MohsCnin NN (1986) Physical Properties of Plant and Animal Materials. Second Edn. Gordon and Breach Science Publishers, New York, MoIto B, BIasco J (2008) Quality evaluation of citrus fruits. Computer vision technology for food quality evaluation, p 247. Molto E, Blasco J, Steinmetz V. Bouncy A, Navarron F. Plertntto (1997) A robotics solution for automatic handling, inspection and ling of fruits and vegetables. Proceedings of the Intemati ual Workshop on Robotics and Automated Macithxery for B iopreduction . BioRob oti£s"97, Grandia, Valencia, Spain. onavar HIM, Alimardani R, Ornid M (2011) Detection of red ripe tomatoes on stem using image processing techniques. Journal of American Science 7(7): 376-379. Monnin M (1994) Machine vision gauging ire a bakery. In: Food processing automation 111. Proceedings of the FPAC Conference (pp. 62-70). St. Joseph, Michigan: ASAE, Morirtoto T Tashi T, Miyata i, Hashimoto Y (2000) Pattern recognition of fruit shape based on the concept of chaos and neural networks. Cornput Electron A is 26(2): 171-16. Morrow CT1 Heiiarn n PH, Sommer 1JJ Tao Y, Vargh se Z (1 99) Automate inspection ofpotatoes, apples, ai I mushrooms. In Proceedings of the luternational Advanced Robotics Prograirime, Aviguori, France: 179--1 S8. Morse BS (2000) Lecture 9: shape description (regions). Brigham Young University Available from: homepagc&inEcd.ac.uk/rb f7 bnlinc/.../rcgion-props-and- moments.pdf. . (accessed 11.03.14). Mukerjec PR and R B Srivastava (1979) increasing the storage life of mango (I angifera indica L.) by lowering the criticl temperature. Prog HoA 10, 3-6 Mukherjee PK,Srivastava RB (1979) Increasing the storage life of mango (Mangifera indica L.) by lowing the critical temperature. POrg. Hcrt. 10: 63-69. Mulchrone XE, Choudhury lit (2004) fitting an ellipse to an arbitrary shape: implication for strain analysis. Journal of Structural Geology 26:143-153. 29 Munek L, Feil C, Oihbans OC (1979) An#lysis of botanical components in cereals and cereals products: a new way of understanding cereal processing. In: Inglett GE, Muk L (eds) In sere is for food and beverages. Academic, Newyork, pp 27-40. Naccache NJ, Shinghal R. (1984) An investigation into the skeietonization approach of HiIditch. Pattern Recommit 17(3): 279-284. Naderi-Boldaji M, Khadivi lthub A, Tabatabaeefar A, Varnamkhasti MG, Zarnarii (2008) Some Physical Properties of Sweet Cherry (Prunus avium L.) Fruit, American-Eurasian J. Agric. & Environ. Sci., 3 (4): 513-520. Nagata M, Cao Q, Bata PM, Shrestha BP, Einosbta 0 (1997) Basic study- on strawberry sorting system in Japan. Imo: 1997 ASAE Annual International. Meeting Technical Papers, Paper No. 973095, ASAE, 2950 Niles Road, St. Joseph, Michigan 49085- 9659, USA, Naglea, Marcus, Intani K, Mahayothee B, Sardsud V, Muller J (2012) Non-destructive mango quality assessment using image processing: inexpensive innovation for the fruit handling industry, Conference on International Research on Food Security, Natural Resource Management and Rural Development organised by: Georg- August Universitat Oöttiigcn $nd University of ssel- itzenhau en Oeniiany September 19-21, 2012. Nair M1 ]ayas DS Bulley MR (1997) Dockage identification in wheat using machine vision. ASAE Annual International Meeting Technical Pap s, Paper No. 973043, ASAE, 2950 Niles Road, St. Joseph. Michigan 49085-9659, USA. Nandani PV, Oommcn BE (2002) Quality parameters of selected mango cultivars. I Food Sci Technol 39: 379-383. Narayana CK, Roy K, Pal RK (1991) Use of high molecular weight ]xigh-density polyethylene I-M-film) in transit of mango (cv. Baneshan). Ac.ta Hort 291:489- 492. Navon E, Miller 0, Averbuch A (2005) Image segmentation based an adaptive local thresholds. Image Vision Comput 23(1): 69-85. Nelson SC (2009) Mango anthracnose ( olletotrchw n g1a sporiodes). Published by College of Tropical Agriculture and Human Resource (CTAHR), Plant Disease, PD-48. Ng HF, Wilcke. WF, Morey RV, Lang JP (1997) Machine vision evaluation of corn kernel rijechauica2 and mould damage. 1n:1997 ASAE Annual International Meeting Technical Papers, Paper No. 973047, ASAE, 2950 Niles Road, St. Joseph. Michigan 49085-9659, USA. Ni B, Paulsen MR, Liao K, Reid JF (19 7a) Design of an automated corn kernel inspection system for machtnc vision. Trans A AB 40(2): 491-497. I 260 Ni B, Paulsen MR. Reid JF (199Th) Size grading of corn kernels with machine vision. in: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973046, ASAE, 2950 Niles Road, St. Josepb, Michigan 49085-9659, USA. Ni 11, Gunasekaran S (1995) A computer vision method for determining the length of cheese shreds. AtTif Intell Rev 12(1-3): 27-37. NIC: National Instrument Corporation (1997) IMA TM Vision for ( Reference Manual. Chapter 3-quantitave analysis, Part Number 321379B-0I, p1-18 Nielsen HM, Piuil W, Munack A, Tantau Hi (199S) Modelling imago processing parameters and consumer aspects for tomato quality grading -In: Mathematical and Control Application in Agriculture and I articulture. Proceedings of the Third ]FAC Worksh6p. Pergamon-Elsevier, Oxford, UK. Njoroge JB, Ninomiya K, Lando N, Toita H (2002), Automated fruit grading system using image processing. Proceedings of the 41 SICE Annual Conference. 2(5-7): 1346-1351. Nutt KP, Hall LD (1999) Advaucc in temperature validation of foods. Trends in Food el Technol 10: 366-374. Novini A (1990) Fundamentals of machine vision component selection. in: Food processing automation II. Proceedings of t e 1990 Conference, Hyatt Regency, Lexington, Kentucky: ASAET p60. Novini A (1995) The latest in vision technology in today's food and beverage container manufacturing industry. In: Food processing automation IV. Proceedings of the 1995 Conference. St. Joseph, Michigan, USA: ASAE. Nowak 3, Mynett K (1985) The effects of growth regulators on the postharvest characteristics of cut lilium prima' infioresoences. Ae&r Hort.16: 167- 109. C' iorman L (1294) Binari tion and nzultithresholding of document images using co me.ctivity. [Trophies Moc1s Image process. 56: 494-5ü6 Omid M, Khojastelmazhand M, Tabatabaeefar A 2010) Estimating volume and mass of citrus Fruits by image processing technique. Journal ]Food Engineering 100(2): 315-321. Omobuwaju TO, Akand EA, Sanni LA (1999) Selected physical, mechanical and aerodynamic properties of African breadfruit (Treculia afficana) seeds. J Food Eugi 40: 241-244. Otsu. N (1979) A threshold selection method from -gray-level histogram. IEEE Transactions on Systems Man and Cybernetics 9: 62-66. Owolarafc OIL, Shotonde 3O (2004) Some physical properties of fresh orko fruit;J Food Engi 63: 299-302. 261 Ozian MM Haciseferogullari H (200 The strawberry , rbukus umedo L. fruits: Chemical composition, physical properties and mineral contents, J Food Engi 78: 1022-1028. Paiiwal J, Borhan MS. Jayas DS (20O3 Classilkation of cereal mains using a flatbed scanner. ASAE paper No. 036103. Pandit RB, Tang J, Mikhayleako G, Liu F (2006) Kinetics of chemical marker M-2 formation in mashed potato-a tool to locate cold spots under microwave sterilization. Journal of Food Engineering, 76(3):353-361. Panigrahi 8, Misra MX. Bern C, Marley S (1 95) Background segmentation and dimensional measurement of corn germplasm. Transactions of the ASAE 38:291- 297, Papamarkos N. Gatos B (1994) A new approach for multilevel threshold selection. Graphics Models Image Process. 5: 357-37O. Park $, Chen YR (1994) Intensified multispectral imaging system for poultry car ass inspection. In Food Processing Automation TIT. Proceedings of the FPAC Conference, St. Joseph, Michigan: ASAF p. 97. _ Park B, Chen YR (1996) Multispectral image co-ocdurrence matrix nalys for poultry carcasses inspection. Transactions of the A AJ 39: 14$5--1491. Park B, Chen YR (2001) Co-occurrence matrix texture features of multi spectral images on poult carcasses. 3 Agric Egg Res 78 [2): 127-139. Park 13, Chen YR, Nguyeu M, Hwang H (1996) Characterising multispectral images of tumorous, bruised, skin-torn, and wholesome poultry carcasses. Transactions of the ASAE 39 (5) 1933: 1941. Park B, Chea YR, Nguyen M, Hwang H (1996) Characterising multispectral images of tumorotjs, bruised, skin-torn, and wholesome poultry carcasses. Trans ASAE 39(5): 1933-1941, Parker JR (1994) Practical Computer Vision using C. John Wiley and Sons Inc., New York, p. 40. Parker JR (1 997) Algorithms for Image Processing and Computer 'Vision, John Wiley & Sons, Inc. p 155-160. Patel HIN, Jain RIC, Joshi MV (201 1a) Fruit detection using improved multiple features- based algorithm. International Journal of Computer Applications 13(2): 1-5. Patel H , Jain IRIS, Joshi MV (201 Ib) Fruit Detection using Improved Multiple Features based Algorithm. International Journal of Computer Applications (0975 — 8887) 13 (2) Paull RE, Chen CC (2004) Mango. In: Gross K.C., Wang C.Y. and Saltveit M. (eds), The Commercial Storage of Fruits, Vegetables and Florist and Nursery Stocks. A draft version of the forthcoming revision to USDA, Agriculture Handbook 66 on the 262 website of USDA, Agricultural Research Service ttp:/lw w.ha.ars.-usda.govlhb 6/ accessed 9 January 2009). Paulus I, De Russchers R, Schrevens E (1997) Use cf image analysis to investigate human quality cla. siftcation of apples. J Arc Eng Res 68(4) 4l-33. Paulus L Schrevcns E (1999) Shape characterisation of new apple cultivars by Fourier expansion of digitai images. J Agric Eng Res 72(2) 113-11S. Pearson T, Toyofuku N (2000) Automated sorting of pistachio nuts with closed shells. Appl Ong Agric 16(l): 91-94. Pearson TC, Slaughter DC (1996) Machine vision detection of early split pistachio nuts. . Trans ASAE 39(3) 1203-1207. Pedresclii F, Leon J. Mery D. Moyano P (2006) Development of a computer vision system to measure the color of potato chips. Food Res Int 39. 1092-1098. Pedreschi F, Mery D,Mendoza F, ingniera JM (2(104) Clnssificatioa of potato chips using pattern recognition. J Food Sci 64: 264 -270. Peli E. 1990. Contrast in complex images. J Opt Soc Am A: 7(10), Persson A-L (1 99) Image analysis of shape and sire of fine aggregates. Engineering Geology 50 (1-2) 177-186. Pesis E, Aharoni D, Aharo Z, Ban-Aria R. Ahrens N. Fuchs Y (2000) Modified atmoshphere and modified humidity packaging alleviates ChilUng . injury symptoms in mango fruit. Postharvest Bio Tech 19: 93-101. Ploetz RC, Zentmyer GA, Nishijirna VAT, Rohrbach KG, Ohr HD (1994) Compendium of Tropical Fruit Diseases. American Phytopathological Society Press. St. Paul, Minnesota. 118 pp. Popenoe J, Hatton. TT, Hsrding PL (1958) Deternunatiou of maturity of hard green Haden and ZilI mangos. Proceeding in American Society Horticulture Science 71: 326-329. .4 Pradeepkumar T (2O0) Management of }iorticultural Crops,Vol.! 1 Horticulture Science Series: In 2 Parts, New India publishing p106. Prusky D (1996) Pathogen quiescence in postharvest diseases, Acnua Revcw of Phytopathology 34: 413 -34. Prusky 1], Ben-Aire R, Guelfat-Reich S (1981) Etiology and histology of Alternaria rat of persimmon frnits, Phytopathology 71: 1124-11 8, Prusky D, Falik E, Kobiler I, Fuchs Y, Zaubernian G, Pesis E, Roth 1, Wel&er A, Akerman M, Ykuticty 0, Waisblurn A, , inan A, Ofek 0, Lavi [J (1997) Hot water bruch anem method for control of postharvest diseases caused by Alternaria rot in nango fruits. 4cIa Borzkulurae 45: 780-785. -14 Prusky D, kicks 1, Kobiler i, Roth 1, Woksler A, ShalomY, FaIik E, Zauberrnan G, Pesis E. Akarman M. Yckutiely 0, Wcisblum A, Rcgev R, Art L (1999), Effect of 263 hot water hurrning, prochloraz fteatmat and axing on The icidenea of black spot ducey caused by Altemaria Altemata in mango cruit, Past Harvest Biology and Technology 15 165-174. Frady D. Kober 1, Ylfiiyara 1, Alan N (2009) Fruit diseases. In: Lit 1.E. (ed), The Mango: Botany, Production and Uses, 2nd edition, pp: 210-230. CAlM publishing, Willlngtbrd, UK. Pun T (1980) A new method for gray-level picture thresholding using the entropy of the histogram. Signal process 2(3): 223-237. Puttaraju TB, Reddy I (1997) Effect of precooliug on the quality of mango cv. `Mallika". Journal of Food Science and Tcchnokgy Mysore 34(1): 24-27, Qiao J, Sasao A. Shibusawa 5, Rondo N, Muriraoto E (2004) Mobile fruit grading robot (Part 1)- development of a robotics system for grading sweet pepper. J. Jpn. Soc. Agric. Mach. 66(2): 113-122. Quevedo R. Carlos LG, Aguilera J 1, Cadoche L (2002) De criptian of food surfaces and micro-structural changes using fractal image texture analysis. Journal of Food Engineering 53: 361-371. l: aghavan cISV, Gariepy Y. Vlgneault C (2003) Coatrolted atmosphere storage. In Encyclopedia of agricultural, food and biological engineering. Marcel Dekker, Inc., New York, p 148. Rajago l (1)97) post-hare t manag mert of n ango - A case sthdy of I sar variety in Junagadh district. Training programme. on post harvest m tnagcment of mango. National lit of Agric Marketing, Jaipur, 4-.December 9-30. Ra5kurnar P. Kaiappan R. Viswanathan R, Parvathi K, Ra pavan GSV, Orsat V (2007) Thin layer drying study on foamed mango pulp. Agricultural Engineering International: the, CRIR Ejournal Manuscript FP 06 024. Vol. IX. March Rashidi M and 5cyti K (2007) Chissification of Fruit Shape in Kiwifruit Applying the Analysis of Outer Dirncnsiors. International Journal of Agriculture & Biology, 9(5): 759-762, Rashidi M, Syfi K. thalami M (2007) Determination of Kiwifruit Volume using Image Processing. J Agri Biol Sd 2: 17-Z2. Rathore HA, Masud T, Samnii S. Soomro AH (2007) Effect of Storage on Physico- hemicai Coiupcsition and Sensory Pop es -of' Mango ( wrigi ra indica L.) Variety Dosehari, Pakistan J Nutrition 6 (2): 143-148, l .avindra I , Goswarni 1' (2008) Comparative peiformnoe of pr=olirg n ethods for the storage of mangoes (Mangi ern indica L. cv. Amtapali). Journal of Food Process Engineering 31(3): 354-371. Ravindra R L Goswami T (2007) Post harvest handling and storage of mangoes-an overview. J Food Sci Technol 44(5): 449-458. . 251 17 zavi S , Pazvar NB (2007) Some Physical and Mechanical Properties of Kiwifruit. In[ Food Eagi 3: 6(3) Read, W (1982) Coniiiertts on two papers in pattern recognition. ThEE Tans on Sirs#. Man, and Cybem 12: 429-430. Reed JN, Crook 8, He W (1995) Harvesting mushroom by robot. In: Elliott TJ (cdl Science and cultivation of edible Fungi, Ba1ktma, Rztterd= pp 3S5-391 Rehkugler GE, Throop JA (1989) Image processing algorithm for apple defect detection. Transactions of the A A E 32 (1): 267-212. Richards A (2006) UV imaging opens new applieations, U , vww.vision-systems com. Ridgway C, Chambers J (1998) Detection of insects inside wheat kernels by N.m. imaging. J Near infrared pest 6: 115-129. Ridge` TW, Calvard S (1 97S) Picture (hresholdiag using an iterative selection method. IEEE Trans on Syst, Man, and Cybern 8(8): 532-630. Rigney MP, Brnsewitz GH, Stnne ML (1996) Peach physical characteristics for orientation. Trans ASAE. 39: 1493-1497. Ritter GX, Wilson IN (1996) Handbook of computer vision algodthnis in image algebra. CRC Press, Riyadi 8, Rahni AAA, Rahni M, Mustafa MIvi, Hussain A (2007) Shape characteristics analysis for papaya size classification, The 5th Studu Conference on Reseach and Development ( OReD), 1 1--12 Dec Rodov VB, Horev YV, Copel A, Aharoni Y, Aharorii N (2002) Modifi edatmosphere packaging improves keeping quality of Chareutais -type melons. Hart Sci 37(6): 950-953. Rojas T, Figueroa 5, Acuzia MC, Rivera F. Diaz De Lean. F, Sandoval MO, Sanchez- Diaz DM, Bosquez E. Madrid R, Kerbel C, Perez L, Vazquez C (1998) Latex chemistry handle export. Revista Ibero mericana Teclmalkgy, Postharvest 1: 26- 31. Roy 5K, Joshi CID (1989) An approach to integrated post harvest handling of mango. Acta Hort 231: 649-661 Roy SK, Pal RIB (1991.) A low cost zero energy cool chamber for short teal storage of mango. Arta Horticulture 291: 519-524. Ruan , Nipg S, Ning A, Jones R, Chen L (1997) Estimation of scabby wheat ineidont rate using machine vision and neural network. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973042, A AE, 2950 Niles Road, St. Joseph, MI 490S5-9659, USA Rush R, Shu N, Luc LQ, Xia C, Chen P, Jones R, Wilcke W, Morey R (2001) EstiroMion of weight percentage of scabby wheat kernels using an automatic 265 machine vision and neural network based system. Transactions of the A Al 44: 983-988. Ruiz LA, Molto E, Juste F, Pla F, Valrtezite R (1996) Location and characterization of the stern-calyx area on oranges by computer vision. J Agiic Eng Res 4(3) 165-172 Ru.s X. (1999) The image processing handbook, 3' edn. Boca Raton: CRC Press, l yatl Al., Lipton WJ (1973) Handling, transportation and storage of fruits and vegetables. Vol. 2, AVE Pub Co.., West port, Connecticut. Saad M, Sadnudi A, 1Rondet F, Cuq B (2011) Morphological characterization of wheat pow&rs, bow to ohaacterize the slap c of particles? Journal of Food Engineering 102: 293-301. Sabato SF, Maria da Silva J, Nunes da Cruz J, Salmieri S1 Rely PR, Lacroix M (2009) Study of physical-cheithc.al and sensorial properties of irradiated Tammy .Atkins mangoes (Mangffera indica L.) in an international consignment. Fond Control 0: 284-288. Sabliov A Boldor D, Keener Iii, Farkas BE (2002) Image Processing Method to Determine Surface Area and Volume ofAxisytrimetric Agricultural Products. mot J Food Prop 5: 641-653. Saha FK, Udupa J (2001) Optimum image dirwho'ding via class uncertainty and region homogeneity. IEEE Trans Pattern Anal Mach latch l 23(7): 689-706. Sahoo P, Wilkins C, Yeager J (1997) Threshold selection using Ren is entropy. Pattern Recognition 30(1): 71-84. Sahoo PK, Soltani 9, Wong AK (1988) A survey of thresh-olding techniques. Computer Visio, Graphics, and IuiageProcessin GIP 41: 233-260. Sapirstein FAD (1995) Quality control in commercial baking: Machine vision inspection of crumb grain in bread and cake products. In: Food processing automation IV Proceedings of the FPAC Conference. ASAE, 2950 Mies Road, St. Joseph, Michigan 49085-9659, USA Sapirstein HD, Kohler 3M. 1999. Effects of sampling and wheat grade on precision and accuracy of k.erne2 feat.rzes determined by digital image analysis. Cereal Chemistry 76:110-1 Ii, Sarkar N, Wolfe RR (1985) Feature extraction techniques for sorting tomatoes by computer vision. Trans ASAE 28(3): 97"79 Sergeant D, Boyle R, Forbes M (1998) Computer visual tracking of poultry. Comput Electron Agdc 21:1-18 Seyedabadi E, Khojastehpour M, Sadmia H, Sniedirad M-R (201 i)Mass modeling of cantaloupe based on geometric attributes: A case study For Tile Magasi and Tile Shabri1 Scientia Horticulturae 130: 54-59 265 Shahin MA, Symons SJ (2001) A machine vision system for grading lentils. Can Binays Eng 7. 7-14. Shahin. MA, Symons SI, Meng AX (2004), Seed Sizing with Image Analysis. A AB paper No. 043121 Shearer SA, Holmes R (1990) Plant identification using coknir co-occurrence matrices. Trans ASAE 33(5): 203 7-2G44 Shearer SA, Payne FA (19 Oa) Color and defect sorting of hell peppers using machine vision. Eng Res 63: 229-236 Shearer SA, Payne FA (1990b) Color and defect sorting. of bell peppers using machine vision. Trans ASAE 33(6): 2045-2050 Shigala 1C, Motonaga Y. Kida T, Matsuo M (2004) Distinguishing damaged and undamaged chaff in rice whole crop silage by image processing. ASAE Annual Meeting. paper No. 043125 Shirai Y (1987) 'Three-dimensional computer vision. Berlin: Springer- crlag. pp. 86-8g. Shimnita K. Miyajima T, Takiyama R (1998) Determination of meat quality by texture analysis. Pattern Recognition Letters 19: 1319-1324. Shukor ARA, l azali M, Omar I] (2000) Post storage respiratory suppression and changes in chemical compositions of mango fruit after storage in low oxygen atmosphere. Acta Hark: 509:467-470. Siddiqui Md W, Dhua RS (2009) Standardization of ethrel treatment for inducing ripening of mango var, International Conference on Horticulture, November 9-12, 8 anglore. Singh BP, Tandoa DYE. Kalra S (1993) Changes in post-harvest quality of mangoes affected by pre-harvest application of calcium salts. Si Hort 54: 211-219 Singh M, Kaur 0, Karr S (2004) Machine vision: an upcoming realm Everyman's science 7 Oct to 7 Nov 42(4): 186-189 Singh N, Delwiche MJ (1994) Machine vision methods for defect sorting stone fruit. Transactions ASAE 7:19891997, Sistler FE (1991) Machine vision techniques for grading and sorting agricultural products. Postharvest News 1nf 2(2): 81-84 Samet R, Ashram A, Abd R , Marruki M, Rini H (2007) Shape characteristics analysis for papaya size classification. The 5th Student Conference on Research and Development- ORaD, Malaysia. Slaughter DC, Obeniai d DM, Thompson JF, Arpaia ML, Margosan bA (2008) on- dcstructive freeze damage detection in oranges using machine vision and ultraviolet fluorescence, Postharvest Bic Technol 4$: 341-346. 267 Slippers B, Johnson GL Crous PW, Coutinho TA, Wingfield BD, Wingfield MJ (2005) Ph logenetic and morphological re-evaluation of the Botryosphaeria species causing diseases of angi; fern rndica in Australia. Mycologia 97: 99-110. Smoot JJ, Segall RIB (1963) Hot water as a postharvest controI of mango anthracnose. Plant Disease Reporter 47, 739-42, Soltani M, Aliruardani K. Omid M (2010) A new inakhernatical modeling of banana fruit and comparison with actual values of dimensional properties, Modem Applied Science 4(5): 104-111 Sonka M, H1avac V, Boyle R (1999) Irna a processing, analysis, and machine vision. California, tad edn_, USA PWS Publishing, p750.770. Somsrivichai J, Gomolmanec S. Boonyakiat D, Utbaibutra J, Beonlong P, Gamma H (1992) Seal Packaging by plastic film as a technique for limiting fungal decay of mangoes. Acta Host 296: 24-32. Spann M, Wilson R (1955) A quad-tree approach to image segmentation which combines statistical and spatial information. Pattern. Recognition 18: 257-269. Spreer W, Muller J (2010) Estimating the mass of mango fruit (Mangifem indica, cv. hok.Anan) from its geometric dimensions by optical measurement. Computers and Electronics in Agriculture 30; 34-30. Steenhoek L Precetti C (2000) Vision sizing of seed corn, In. 2000 ASAE Annival 1ntemational Meeting, Paper No. 003095 ASAE, 2950 Niles Road, St. Joseph, MI 49085-9659, U A Steinmetz , Roger 3M, Molto E, Blasco J (1-999) On-line fusion of colour camera and strawberry using machine vision (part 2): development of sorting system with direction and judgement fuuctio for stiuwbtrry (Akihirne variet). J Jpa Soc Agri Machinery 62(2):101-1 10 Steward RI1, Tian LF, Nettleton DS Tang L (2004) Reduced-Dimension Clustering for Vegetation Segmentation. Transactions of the ASAF, 47(2); 609-616. Stone ML, Kranzler GA (1992) Image based ground velocity measurement, Trans A AI.35(5):1729-1734 Straten BU, Oshu se SA (1994) The effect of cold storage at 4 of 8 degree on the rate of quality deterioration of ripe sensation mango fruit. South African mango growers association year hook 14: 34-36. Sun D-W (2000) Inspecting pizza topping percentage and distribution by a computer vision method. J Food Eng 4-4(4): 245-249. San D (2004) Computer vision - an objective, rapid and noncontact quality evaluation tool for the food industry. J food Eng 61; l-2 Sun D-W (2008) Computer vision technology for food quality evaluations Food Science and Technology, International Series, Ist edit., Elsevier Inc. 263 Sun D-1 . Brosnan T (2003) Pizza quality evaluation using computer vision-Part 1. Pizza base and sauce spread. J Food Eng 57 8 I-89. Sun X Litchfield JB# Schmidt SI (1993) Temperature mapping in a model food get using magnetic resonance imaging. J Food Sd 58: 168-172, 181. Sun XZ, Schmidt 5J, Litchfield JB (1994) Temperature mapping in a potato using half fourier transform MRI of diffusion. J Food Process P.ngg 17: 423437, Taba#abaeefar A, l ujabipour A (2005) Modeling the mass of apples by geometrical attributes. Sci.Hortic. 105: 373-382. Tendon DI.-, Kalra 8K (1986) Studios as developing mango fruits to asses maturity. IrLdian J Hort 43: 51-55. Tandon DID, Kara SK (1986) Studies on developing mango fruits to asses maturity, Indian Journal of Horticulture 43: 51-55. Tau Y (1996b) Methods and Apparatus for Sorting Objects Including Stable Color Ttnsformatjon. United States Patel t 5 533 628. Tao Y, Deck SJ-1, Morrow CT, Heinemann PH, Sommer HJ, Cole RH (1991) RuIe- auper+ sed automated machine -ision inspection of produce. St Joseph MI, ASAE Paper: 91-7003. Tao ", Heineman.n PH, arghose Z, Morrow CT, Sommer H.1 III (l995a) Machine vision farecloux inspection of potatoes and apples. Trans ASAE 38(5): 1555-1561, Tao Y, Morrow CT, Heinemann PH, Soininer HJ (1995b) Fourier based separation techniques for shape grading of potatoes using machine vision. Trans ASAE 38(3): 949-957. Tao Y, Wen Z (2003) Item defect detection apparatus and method. US Patent, Publication No. US6610953 BI Tarbell 1(A, Reid JF (1991) A computer vision system for characterising cori growth and development, Trans ASAE 34(5): 2245-2249. Teoh CC, Syaifiudi AR Mohd (2007) Image processing and analysis techniques for estimating weight of Chokanan mangoes. J. Trop. Agric. and Fd. Sc. 35(1) 183- 190. Teoh CC, Sysifudin ARM (2006) Use of image analysis for grading sing of mango. Arta Horticulturae, Numb 710: 485-490 1 Thakur M, Quan C, Tay CJ (2007) Surface profiling using fringe projection technique based an Lua effect. Opt Laser T hnol 39: 453-459. 'Thompson IF, Mitchell FCs and Kaiuiie RF (2002) Coating horticultural commodities. In. Postharvest technology ofhwrticuitural crops, Third edition, Kader AA (Tech. Ed). University of California, Division of Agriculture and Natural Resouxcos, X'ub&ation 3311: 97— 112. Throop JA, Anoshan.sley DJ (1993) Investigation of texture auatysis features to apple bruises, ntemational winter meeting. ASAP, presentation paper no, 933527. Throop JA, Aneshansley DJ, Upchurch BL (1995) An image processing algozithm to find new and old bruises. Appl Erg Agri l 1: 751-757. Tillett ND, Hague T (1999) Computer visiun based hoc guidance for cereal -an initial trial.. Agric E g Res 74(3) 225-236, TilIutt RD (1990) Image analysis for agicultura1 processes. Silane Research 1nM tote. Division Note DID 1585. ' immermans AiM (1998) Computer vision system for on-line sorting of pot plants based on learning techniques. Acta Hart 421: 91-98. Tobias OJ, Sears R (2002) Image segmentation by histogram #hresholding using fuziy sets. IEEE Trans an Image Process 11(12):1457-1465. Tamasi C, Manduchi R (1998) Bilateral filtering for gray and color images, In: Proceedings of IEEE International Conference on Computer Vision (pp. 839- 846), IEEE Bombay, India. Topaz A, Topakci M, Canakci M, Aldnci I, Ozdemir F (2005) Physical and nutritional properties of four orange varieties. J Food Engi 66: 519-523. Tsai WH (1985) Moment preserving thresholding: a new approach. Comput Vision, graphics, and Image process 29(3): 377-393. Ueda M. Sasaki K. Utsunomiya N. Inaba K, Shimabayashi Y (20 C)!) Effect of temperature and time on some properties during storage of mango fruit (A6ngg era inc ca L. Irwin') cultured in plastic house. Journal of the Japanese Society for Food Science and Techto1ogy 48(5): 349-355. [nay 0, Gosselin B (2 )07) Stem and calyx recognition on 'Jonagold' apples by pattern recognition. Journal of Food Engineering 78: 597--605 Upadhyay NP, Trpathi BM (1985) Post harvest changes during storage and ripening urject mango (Mangifera Indica L.) fruit Prog. Hart. 17: 25-21. Upchurch BL, Affeldt HA, Hruschka WR, Norris KR, Throop JA (1990) Spectrophatometric study of bruises on whole, Red Delicious' apples. Trams A AE 33 (2) 585-589. USDA (2001) Nutrient Database for Standard Reference, Release 14 July 2001. Uthaisornbut P (1996) Detecting Defects In Cherries Using Machine Vision. Thesis, published online, citoseerx.ist.psu.edu/viewdocldownload?doi=10.1.1.34.3500. Valero D. Perez-Vicente A, Martinez-Romero D, Castillo S Gui11et F, errino M (2002) Plum storability improved after calcium and heat postharvest treatments: Role of pulyumines. J Fogel Sci 67: 2571-2575, Van Eck 1W, van der Hcijden iWAM, Polder G (1998) Accurate measurement of size and shape of cucumber fruits with image analysis. I A ric Engi Res 70:335-343. 270 Van Henten El, Hemming J, Van Tuijl BA], Kornet JG, Meuieman J,, Hnntsema J, VanOs EA (2002) An Autonomous Robot for Harvesting Cucambers in Greenhouses. Autonomous Robots 13: 241-258. Vandevoorea JO, Folder (3, van der Heijden G (1992) Identification of mushroom cui ivars using image analysis. Transactions of the A AE 35: 347-350 VanLinden V, Veieycku R, Bravo C, Ramon H, Baerdemaeker JD (2003) Detcctit n technique for tomato bruise damage by thermal imaging. In Proceedings of Post- harvest Unlimited, 389-394. Leuven, Belgium: I H . Varith J, Hyde GM, Baritelle AL, Fellman 1K, Sattabcrngkot T (2003) Non- contact bruise detection in apples by thermal imaging. Innovative Food Sci Emerging Technol 4: 11-218. Vaysse P. Grenier G, I.avialle 0. Henry , Khay- hbat 5, Gerrnain C, Da Costa JP (20G) Image processing as a tool for quality assessment of fruits in bulk shipping bins. Information and Technology for Sustainable Fruit and Vegetable Production FRLIT1C 05, 12, Montpellier, France 381, Vazquez-Caicedo AL, Neidhart 5, Carlo R (20(14) Postharvest ripening behavior of nine Thai mango cuItivars and their suitability for industrial applications, Acts Horticuitu ac 645: 617-625. VazquezCaicedo AL, Sruamsiri P, Carle R, Neidhart S (2005) Accumulation of all- 1ra-carotene and its 9-cis and 13-cis stereoisc>mcrs during postharvest ripening of nlae Thai mango eultivars. Journal of Agicultural and Food Chemistry 53: 4827-4835. Vincent I,, Soille P (1991) Waters hed in digital. space: An efficient algorithm based on immersion simulations. IEEE Transactions on Pattern Analysis and Machine Intelligence 13(s): 583-593. Vizhanyo T, Felfo ldi J (2000) Enhancing colour differences in images of diseased mushrooms. Comput Electron Agric 26:187-198 Vizhanyo T. Tillett RD (1998) Analysis of mushroom spectral characteristics, Proceedings of AgEng'98 International Conference on Agricultural Engineering, Oslo, 24-27 August, Paper No. 98-F08019981-7 Vnrsavus K, Kdebek I, Selli S (2006) A study on ' some c•hemie I and physicomechanical properties of three sweet cherry (Prunus aviwn L.) varieties in Turkey, J Food Bag 74: 568-575 Wan YN, Lin CM, Chiou JF (2000) Adaptive classification method for an automatic grain quality inspection system using machine vision and neural network. In: 2000 A AE Annual International Meeting, Paper No. 00 094 A AE, 2950 Niles Road, Si, Joseph, Ml 49085-9659, USA. Wang Ii-H, Sun D-W (2001) Evaluation of the functional properties of cheddar cheese using computer vision method. I Food Eng 49 (.1): 47-51 Z71 Wang H-H, Sun D- (2002) Correlation between cheese meltability determined with a computer vision method and with Arnott and Schreiber. Journal of Food Science, 67(2) '145-749. Wang H-H, Sun D- (2002a) Melting characteristics of cheese: analysis of effects of cooking conditirns using computer vision techniques. I FoQd Eng 52(3): 79- 284, Wang Q. Ch Z, Zhao (2002b) Image thresholding byamaxunizing of noafuzzi.aess of the 1) grayscale. histogram. Computer Image and Vision Understanding S5: 100- l16. Wang T , Nguang SK (2007) Low cost sensor for va1r~ni and surface computation of axi-symmetric agricultural products. J of Food Engg 79: 870-877. . Wang Y.-C, Chop N' (1996) Automatic segmentation of touching rice kernels with an active contour model. Transaction of the ASAE, 47(5): 1803-1811. Wardlaw CW, Leonard ER (1936) Studies in tropical fruits. I. Preliminary observations on some aspects of development, ripening & senescence with special reference to respiration, Awn. Botany (London) 50: 621-653. Wardowsld VHF, Miller WM, irierson W (1997) Separation and grading of frc e.- damaged citrus fruits. Circular 372, In$t'stute of Food and Agricultural Sciences, University of Florida, Wakar DP, Dhemre JIB (2005) FfI`ect of preconling on extending thepostharvest life of kesar mango fruits. Indian J agric Res, 39(1): 42-6. Wen Z. Tao Y (1998) Method of dual-camera NIRJMIR image for fruit sorting. ASAE paper 983043. St_ Joseph, MI White KW, Sellers ICJ (1994) Foreign materials sorting by innovative real time coIor signatures, in Food Processing Automation III-Proceedings of the FP,AC Conference, ASAE, St. Joseph, Michigan, p, 29, Whiltakec Al), Park B, Thane BR, Miller RK, Savell 1W (1992) Principles of ultrasound and measurement of intramuscular fat. J Animal Sci 70: 942-952. Wilhoit JH, Auburn AL, Klutz LJ, Fly 1W., South OB (1994) P -based multiple camera machine vision systems for pine seedling measurements. App! Brg Agric 10(6): 841-847. Williams AL, Heineman PH (1998) Detection, tricking, and selection Of mushrooms for harvest using image analysis. In ASAE Annual International Meeting, Orlando, FL, USA, 12-16 July 1998. ASAE Paper no. 9304I. Wills R, Lee T, Graham D, Meglasson W, Hall E (1981) PostbarvesL An introduction to the physiology and handling .of fruit and vegetables, Granada. Wooten JR. White J , 7howasson JA, Thompson PG (2000) Yield and quality monitor for aweetpotato with machine visi . It 2000 ASAE Annual International Meeting, Per No.001123, ASAE, 2950 Niles Road, St. Joseph, M[ 49085-9659! USA. Wycislo AP, Clark JR, Karcher DE (2008) Fruit shape analysis of visit using digital photography. Hort. Sri. 43(3): 677-680, Xie W, Paulsen MR (1997) Machine vision detection of tetrazalium in corn. In: 1997 ASAE Annual Intcmational Meeting Technical Papers, Paper No. 973044, ASAE, 2950 Niles Road, St. Joseph, MT 49085-9659, USA. Xing J, Bravo C, Moshou D, Raman H, De Baerdemaeker J (2006) Bruise detection on Golden Delicious'- apples by VIS1NIIR spectroscopy, Comp Electron Agri 52: 11-- 20. Yahia EM (2005) Mango Mangifema Indira L), Ch 22. Postharvest Biology and Technology of Tropical and Subtropical Fruits, edited by E Yahia. Woodhead Publishing Limited, 2011. p492-565. Yahia EM, Vazquez-Moreno L (1993) Tolerance and responses of mango to insecticidal oxygen and carbon dioxide atmospheres. Lebensm-bliss Technol 26: 42-48. Yam K ,, Papadakis SE (2004) A simple digital imaging method for measuring and analyzing colour of food surfaces. J Food Fngi 61: 137-142. I'an F, Zhang H, Kube CR. 2005. A multistage adoptive thresholding method. Pattern Recognition Lett, 26(8): 1183-1191, Yang Q (1993) Finding Stalks and Calyx of Apples Using Structured Lighting. Computer Electron Agri 8: 31-42 Yang Q (1994) An approach to apple surface feature detection by machine vision. Comput and Electron Agric 11: 249-264. Yang Q (1996) Apple stem and calyx identification with machine, vision. J Agric Eng lies 63(3): 229-236 Yang Q, Marchant JA (1995) Accurate blemish detection with active contour models. Comput Electron Agric 14(1): 77-89. Yang Q, Tillett RD (1994) Apple surface feature for blemish detection, Proceedings of workshop COST 94-System and Operation for Post-harvest Quality, September 1993, Leuven, Belgium, pp, 93-107. Yin H. Panigrahi S (1997) Image processing techniques fir interval texture evaluation of French fries. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973127, ASAE, 2950 Niles Road, St, Joseph, III 49085-9659, USA. Ying Y, Jing H, Tao Y, Zhang N (2003) Detecting stern and shape of pears using Fourier transformation and an artificial neural network. Transactions of the ASAE 46(1):157-162. Ying YB, Xu ZG, Fu XP, Liu YD (2004) Nondestructive maturity detection of Citrus with computer vision. In Proceedings of the Society of Photo-optical 273 Instrumentation Engineers (PIE), 5271, Monitoring Food Safety, Agricultur , and plant Health, p 97-107. Yong lie, Yun Zhang. Annia G. Pereira, Ant hus H. Gomez & Jun Wang (2005) Nos de tr wlive Determinat[kn of Tomato Fruit Quality Characteristics Using VisIN]IR Spectroscopy Technique, International Journal of Information Technology I1: 11, Young T (1802) On the theory of light and colors. Philosophical Trans. of the Royal Societ' of Lundan 9220-11. Youswf MIM, Abou-Giribia HA, El-Haddad N (20051- Applications of digital image analysis to food quality analysis assessment: an overview. Alex Fd Sci Technol 2(l):41-50. Zafed NS Khaf'agy SAAJ Saleh MA (2007) Evaluation of some mango species by foot characters and fingerprint. Res J Agric Biol Sci 3(4: 316-320. Zayas L Pomeranz Y, Lai FS (1989'p Discrimination of wheat and non wheat components in grain samples by image analysis. Cereal Chem 66(3):233-237; Zayas I , Martin CR. Steele JL, Katsevich A (1996) Wheat classification using image analysis and crush force parameters. Trans ASAE 39(6): 199-2204 Zhang L, .)in Y, Lin L, Li J, Du Y (200) The Comparison of CCI) and CMOS Image Sensors. International Conference on Optical Instruments and Technology: Advanced Sensor Technologies and Applications, edited by Anbo Wang, YanRiao Liao, AiGuo Snug, Yu1dhro Ishii, Xudong Fan, Proceedings of SPIE \IOL 7157, 71570T-1. Zhang T, Deng J (1999) Application of computer vision to detection bruising on pears. Trans Chin Sec Agric Eng 15(1): 205-209. Zhang ZQ, Li XP, Hong lU, Ji ZL (2000) Alleviation of chilling injury of mango fruits in cold storage by using exogenous putrescine. Fujian J Ag6c. Sci 1532-36. Zheng C, Sun D-W . (2008a) Image segmentation techniques. In: Sun D- , editor. Computer vision technology for food quality evaluation, l eda. Elsevier, Inc, pp 37-56. Zheng C, Sun D-W (008b) Object measurement methods. [n Sun D-W, editor. Computer vision technology for food quality evaluation, I" edn. Elsevier, Inc, pp 57-50. Zheng C, Sun DW, Zheng L (2006b) Recent developments and applications of image features for food quality evaluation and inspection-a review. Trends in Food Science and Technology 17: 642-655. eng C, Sun OW, Theng L (2006c) Estimating shrinkage of large cooked beef joints during air bast cooling by computer vision. J Food Engg 72: 56-62. 274 Zheig C, Sun D-W, Zheng L. (2006a) Receat development of image texture for evaluation of food qualities.a review. Trends in Food Science and Teclmolo + 17: 113-128. Zion B, Shklyar A, Karp1us 1 (1999) Sorting fish by computer vision. Computers and Electronics in Agric 23:175-187. • Zion B, Shklyar A, Karplus 1(2000) In vivo fish sorting by computer vision. Aquacultural Eng 22:165-179. Zuech N (2000) Vnderstand.ing and applying machine vision. 2nd ed., M cet Dekker, Inc. p0.-. - Zueh N (2004) Machinc vision and lighting. Machine vision on line htrp;./f w.machinevisionontino.org/public/artic1es/archivnetails.cfnx?id=1430, (accessed on 5 NOV. 2010). . ppend -I A-1,1 Anova of the dependent variable length or major diameter of niango.fruits of Chausa and Dasliehar/ Some Type III Sum df Mean F Sig. of Squares _ _ Square _ Corrected Model. 24202.433 11 2200.221 24.156 0,000 intercept 2788106.183 1 2788106.183 30610.863 0.000 Variety (V) 22144.737 3 7381.579 91.043 0.000 Harvest Stage IIS) 74.405 2 37.203 0.408 0.665 VHS* 1484,531 6 247.422 2.716 0.04 Error 26504,934 291 91,05 Total 3289836,460 303 Corrected Total 50707.367 302 A-1.2 Anova of the dependent variable breadth or intermediate diameter of mango fruits of Chausa and Das iehari Source Type III Sum df Wan Square F Sig. of Squares Corrected Model 8804.122 11 800.375 32,626 0.000 Intercept 1047 87.387 1 1047687.357 42706.844 0.000 Variety (V) 8111.452 3 2703,817 110.216 0.000 Harvest Stage (HS) 552.144 2 276.072 11.254 0.000 VHS¢ 475 233 i 79.205 1+229 0.004 Error 7138.833 291 24.532 Total 1158498.730 303 Corrected Total 15942.955 302 276 A-1.3 Anova of the dependent variable thickness or minor diameter of mango fruits of Chausci and Dashehari Source Type m Sam df Mean F Sig. - of Squares Square Corrected Model 8399.968 Ii 763.633 39.373 0.000 Intercept 862710.837 1 862710.837 44481.456 0.000 Variety (V) 6752.777 3 2250.926 116.058 0.000 Harvest Stage (HS) 1718.047 2 859.024 44.291 0.000 V*HS 794.979 6 132.497 6.832 0.000 Error 5643.899 291 19,395 TOW 947459.520 -303 Corrected Total 14043.867 302 A-1.4 Anova of the dependent variable weight of mango fruits of Chausa and s/whari cultivar5 Source Type III Sum df Mean F Sig. of Squares Square Corrected Model 734025.701 11 66729.609 36.103 0.000 Intercept 11846120.537 1 11846121154 6409.206 0.000 Variety ( 615554.626 3 205184.875 111.013 0.000 Harvest Stage (H) 117433.146 2 58716.573 31368 0.000 V*II5 79065.487 6 13177,581 7.130 0.000 Error 537854,614 291 1848.298 Total 14065999.374 303 Corrected ToW 1271880.315 302 277 A-1.5 Anova of the dependent variable weight of mango fruits of Chausa and Dashthari cultivars Source Type III Sum df Mean Square F Sig. of Squares Corrected Model '730444.458 - 11 66404.042 37.590 0.000 Intercept 12220000.068 1 12220000.068 6917.546 0.000 Variety (V) 63 1 126.211 3 210375.404 119.090 0.000 Harvest Stage [H } 101051.185 2 50525.593 25.602 0.000 V*HS 63492.606 6 10582.101 5,990 0.000 Error 514058.018 291 1766.522 Total l4556450000 303 Corrected Total 1244502.475 302 A.-1.A Anova zf the dependut variable volume of ma go f its of Chausa and Das r ari eultivars Source Type III Sum df Mean Square F Sig. of Squares Corrected Model 80l021 11 728.256 25.055 0.000 Intercept 1457709,059 1 1457709.059 30151.988 0.000 Variety ( 7159.078 3 2386.359 82.102 0.000 Harvest Stage (11 5) 555.811 2 277.905 9.56I 0.000 *I'IS 64.853 i 114.142 3.927 0.001 Error 8458.156 291 29.066 Total 1650345.680 303 Corrected Total 16468.977 302 278 A-1.7 Anova of the dependent variable arithmetic mean diameters of mango fruits of Chwsa and Dashehari cultivars Source Type U1 Sum df Man Square F Sig. r Squares Corrected Model 8010.821 11 725.256 25.055 0.600 1nterr~ept 145709.059 1 1457709.059 50151.998 0.000 5 Variety [V) 7159.079 3 2386.359 82.102 0.000 H .rvcst Stage (1-IS) 555.811 2 277.905 9.561 0.000 Vic 684.853 6 114.142 3.927 0.001 Error 8458.156 291 29.066 Total 1650345.680 303 Corrected Total 15468.977 302 A-L.8 Anova of the dependent varlabie geometric mean diameters of mango fufts of Chaa~sa and Dashehari cultivars Source Type III Sum df Mean Square F Sig. of Squares Corrected Model 8119,410 11. 735.128 2.8,706 0,000 Intercept 1354463.230 1 I34463.230 5261 .005 0.000 Variety (Y) . 7208.665 3 2402.888 93.448 OMDU Harvest Stage (IfS) 746.788 2 373.394 14,521 0,000 *i1 $66.164 6 111,027 4.315 0.000 Error 7482.653 291 25 714 Total 152019$.844 303 Corrected Total 15602 064 302 279 A-1.9 Anova of the dependent variable sphericity of mango fruits of Chausa and Dashehari cult vats Source Type III Sum df Mean Square F 1g. of Squares Correetcd Model 0399 11 7.264E-02 53.696 0.000 Intercept 133.853 1 133.853 98947.779 0.000 Variety( 0.717 3 0.239 176.730 ' 0.000 Harvest Stage (HS) 5,999E-02 2 3MOOE-02 22.174 0.000 V*HS L365-02 6 3,108E-03 2.298 0.035 Error 0.394 291 1,353E-03 Total 143.887 303 Corrected Total 1,193 302 4-1.10 Anova of the dependent variable aspect ratio of mango fruits of Chntsa and ))ashchari cultivars Source Type I31 Seim df Mean Square F Sig. of Squares Corrected Model 1.251 11 .114 48.272 0,000 Intercept 104.188 ! 14.198 44232.668 0.400 Variety (V} 1.I76 3 .392 166.370 0.000 Harvest Stage (H) 5.124E-02 2 2.562E-02 10.878 0.000 *flj5 1.663 E-02 6 2.771 &03 1.177 0.3 19 'Error D.63. 291 2.355E-03 TOtaW 1RL900 303 Corrected Total 1.936 302 280 A-1.11 Anova of the &,pendent variable 'ASS of mango fruits of Chrn,sa and L)ashehari cultivars Source Type LII Sam df Mean Square F Sig. of Squares _ _ Corrected Model 5191.333 41 126.618 148.185 0.000 Intercept 52052.220 1 52052.220 60918.198 0.000 Variety (} 1238.586 3 412.862 483.1$4 0.000 Harvest Stage (HS) 215.062 2 107.531 125.547 0.000 Storage Periods (SP) 3451.E 16 5 690.243 807 811 0.000 V * 118 135.543 6 22.590 26.438 0.000 V * P 91341 15 6.089 7.127 0.000 HS * SP 59.585 10 5.959 6.973 0.000 Error 148.676 174 0.854 Total 57392.230 216 Corrected Total 5340.010 215 A-x.12 Anova of the dependent variabl.e acidity of mango fruits of C'hau$a and Dasheari cultivars Source Type Ill Sum df Mean Square F S1g, of.Sguares Corrected Model 133.980 41 3.268 21.576 0.000 Intercept 221.676 1 221,676 1463.623 0.000 Variety(V) 14,491 3 4.830 31.892 0.000 Harvest Stage (H) 15.155 2 7.578 50.031 0.000 Storage Periods (SP) 75.902 5 15.180 100.229 0.000 V * HS 5.321 6 0.887 5.855 4.000 V* SP - 12.816 15 0.854 5.641 0.000 HS * SF 10.295 10 1.029 6.797 0.000 Error 26.354 174 0.151 Total 382.009 216 Corrected Total 160.333 215 281 A-1.14 Anova of the dependent variable specific gravity of mango fruits of Chausa and Dasheharri cultivars Source Type III Sum df Mean Square F Sig. of Squares Corrected Mode! 30.947 41 .755 23.209 0.000 Intercept 58,531 1 58.531 1799.731 0.000 Variety ( 3.939 3 1.313 40377 0.000 Harvest Stage (HS) .644 2 .322 9.895 0.000 Storage Periods (F) 21.237 5 4.247 130.602 0.000 V*HS 2.524 6 0.421 12,933 0.000 V* P 1.987 15 0,132 4,074 0.000 H * P 0.616 10 6.157E-02 1.893 0.049 Error 5.659 174 3.252E-02 Total 95.136 216 Corrected Total 36.606 215 A-1.14 Anova of the dependent variable specific gravity of mango fruits of Chausa and Dashehari cultivars Source Type HI Sum df Mean Square F Sig. o1 Squares Corrected Model 0+561 41 1.369E-02 3,247 0.000 Intercept 214.772 1 214.772 50953.193 0.000 Variety (V) 9.553E-OZ 3 3.1 94E-02 7.579 0,000 Harvest Stage (HS) 6.263E-02 2 3.1311-02 7,429 0.001 Storage Periods ( F) 251 5 5.016-02 11,901 0.000 V*HS 3.587E-02 6 5.978E-03 1.18 0,210 V * SF 4,655E-02 15 3.103E-03 0.736 0.745 HS * SP 6.948E-02 10 6.98E-03 L648 0,097 Error 0.733 174 4.215E-03 Total 216.061 216 Corrected Total L295 215 282 ri A-lAS Anova of the dependent variable °L' col-or values of mzngo fruits of Chaua and . tehar i cultivars Source Type III Suns df Mean Square F Sig. _ of Sgres _ _ Corrected Model 7976.203 41 194.542 33.073 0.000 Into -cept 502214.615 1 502214.615 85379.641 0.000 Variety (V) 1364.382 3 454.794 77.317 0.000 Harvest Stage (ES) 798.937 2 399.468 67.911 0.000 Storage Periods (SP) 3740,270 5 748.054 127.172 0.000 V * IfS 1787.205 6 297.867 50.639 0.000 V * SP 118.889 15 7.926 1.347 0.179 H * SP 16-6.519 10 16.652 2.831 0.003 Error 1023,504 174 5.882 Tota' 511214.321 216 Corrected Total 8999,706 215 A.-1.J6 Anova of the dependent variable 'a1 color values of mango fruits of Chau.90 and Dushchari cu Itivars Source Type Ill Sum of df Mean Square F Sig. Squares Corrected Model 2329,193 41 56.810 19.855 0.000 Intercept 9288.846 1 9288.846 3246.378 0 000 Variety (V) 231.537 3 77.179 26,974 0.000 Harvwst Stage (1-IS) 108.969 2 54.484 19.042 0.000 Storage.Periods (P) 1432.924 5 286.585 100.159 0.000 V*HS 61.998 6 10.333 3.611 0.002 V * P 326,401 15 21.760 7.605 0.000 H * P 167,364 10 16.736 5.849 0.000 Error 497.865 174 2,861 Total 12115.905 216 Corrected Total 2827.059 215 283 A-I.17 Anova of the, dependent variable `b' color values of mango fruits of Causc and Das/eharj cultivars Source Type III Sum dl Mean Square F Sig. of Squares Corrected Model 5771,356 41 140,765 9.954 0.000 Intercept 1857D8.812 1 185708.812 12999 864 0.000 Variety (1.J) 1165.579 3 388.526 27.197 0.000 Harvest Stage (H) 2214.432 2 1107.216 77.507 0.000 Storage Periods (P) 194374 5 38.875 2.721 0.021 V*liS 602.488 6 100.415 7.029 0.000 V* SP 457.102 15 30.473 2.133 0.010 HS * P 1137.381 10 113.732 7.962 .000 Error 2485.667 174 14.285 To#aI 193965.835 216 Corrected Total 8257.023 215 -2d. A1p,i thrn d M t.thN!W ioitwarc for image {M1 ' air d frujt.q from m~noarirrnc camera __` I•L a 111 P~~~f~~11•• —I-.L •'•~ _____~ 4i`i liiini' 'FI 11 ++'y p T II_ h 1111 T T' •'T. •I {) C ia[ Ln br~w~ d & s l «d (nhaE fbrighm~ L..f L.'-,• D riHD .,1 Y~~luli~ I 5ML iW. 6&111 L.Y_. L~ LI- ..y.~5 .5• '~. Ili J it ~ ii ,,L ,I.i ., rJ. .~. - ,'R.'.M. v1- n ITI■ W. J ,, .. (d} i]rehc 9baar! ~f ILtl ()?ie1 E[11ul1 ;4tIM (IUT) " ' L1~~1+51r M1'u. :-, I11 I'. 11,1 .__ I i f '1 7 J- ' • ~5 '•' ' [~@ aval 1 , '• I ' re,,w -r rl,l w •— L'~ J Li L 1 C6dl1tC[llCQk-dLIDnñ3 ( 4 ( P ¢iicl utpul) (h) Edg deL !k is ouk1 ii d imae 25 VAILr Aipuiillti) d i'iptini o i LthVIE' sil!w& f f !1(L 1Dhoi frui[s f roan till ±J(1.t 3iThtLI Oin r aid & srJct! (r) I:! mjwli+ thh (d) Tiic h I obJ' lI1T~ % I,.1 -„._fir , ; r „ I'll U T1 ,I• h Edg Ee o rrutl ind ilnagc ail hip ur mrukJimrn:uuS ~ ° ai(pu{] 286 Appendix-2b A-lb.11 A prichn d ripti i OL I.abVI. \V Bohm- ibr iüt f tad frui1 Earn colur (IA A) CYS 1TTTLV ■ L1 Y ~. .(j • I I - - 55 I A ' I 1 In J 1'I IA, t f - • I• L....• .I 1i I I rl I' 1V .1 1,1 rl IF Iui" .li ••11 Ir'1 'I f'r LJ' 11 A. I11 L. J ' (i) Qriori t ii i c hr cd & ) tri' of hri h1n () lid DID p] • A L _£ IJ t III , -1_ I (rI) TIir finIl1 I1b1 L"t 16 PiNd nuuIir time ~ti5~rlll, L.-LII I_ H r' 11 iIL u ' _ }nvuiJJ[ - I J () J'9i~idl ii91yik ( QulEru1) (hJ dclat!i :•outlir Iinmge (i) n iii1 21 A-2U. Al tithm &cFL1oi1 On L bVIL~W th'ütt lbr ima o f Aat a ri rti from c1nr (R IE) CVS (a) Odgin ] ]rasp brwd aadd scl-6cbr%l fib) Euhg«amal of brigh os } Pil Ioutpat) ( UU1P uli,u: ucili d imu' i} Masuj au~aiit-d~ nsi ~~ ('~ nutty 23 ppeiidix'lc A•2[.1. Aid ithm szri~4rn rm I~h V1B" ! softer re t r of C i fruit 6mrn N CVI -LY~n■' aver ~~} ~1 •- - - -_:-- ' rl•;-- : J.LJ.Ll1I u— - (j Eime E of brigh s of btu plane I r-Ir• ' • L' VI Ii I11 II ~kl•. Pic ([' o¢1pu) (h~ Lc g c li ~: oullin l im ip () M Lii~flt u77ri ici I t 2$~ A-h.1. r ]~«iL~u~i diripLii on L~,V1k \ Ow rc firin 'I' ukE cr fruits in V CYS ~LLLI ~ ~.~~.. _.,, .. MRS 'll I•. F•1~ ~rTix I I I nl III _~I~INhI~ _• •r I I r - •••. 'I Y■ I~ I '/Y I.11 111; 1 ~J/ 'll~ I'' '• -= r~ iy LII'•. „ -'~y '~~^•~.- --~11 ---. - I— YYff YN IYT 11• •~ I I ( of bri ~I7i ne 1 of 111iie Liitia (a) Imaoc br ud aI d I X!od li1S•tri I Sii I pl i K1c 1ii7 I te, ' II ".I — ,YI,f 1I1 5 t a r r• (o Particle r autpit) Ed d t aoa, k image i0 MmwT=['sliuiisiu15 (lug 4rtprt) 290 AppendIx•1d A-k I. Al rnithm dccGrip[iU L by I i sDftwa for hn f Chliq fruib ftam ilk CVS -,'-;te—. _,. '.,--- y'' I M W 1 J Y II • 11 L J V_ 1_ 1 i 1, i V w RI M nu ?I- 'V I — — Y _ - lsnlwn mcnt of i m o hip[ i~ ~If dn~ [ i1 ' pls~ cOAC[i pn 1 '~I NI1 Iy - I 1. I u • ,1,I, [ii: r,r 5 (d) !In" uhjG( '—I r, T I ______r .______ r__ ...u. I. .1 ' ~ . I ~ Y • _ ■ 1■ •I r ? i1k'1C ii1Cysi"a (r1 01II!Ii) h) L111 Cd l Q3t 111g (i) 21 A-2dI2l, Ig ihhin dciptIo1 on L VIEW iftwun. fir irn~ 9 of Dai) hari fruit PrimNER CV {a} hap 'gcd aiand x] *d (b~ F.nhanctcjA of'lxiphwi 1hiuc (G] Lquali~etian oF pixcl (L]]Tl () Prickanal)lis (r'rnut fit) E F ~lair:th~ Vu~i O lump nutput) 292 Appeiidh-3 A.11, The it fi values, st f dird JviJiiou [S.D.) ani yang (rein, nix) of oiu un,dnc altrihuth9 f ma a iIII (C qty, i Dd,Q~reir ukivr) ariilyzd by lr rini i s Ing mo 11 o!ne Amer' use CVSI - — -~ — — --- — — — — — Mouochrorile camera bud CY~ —. - Puraaai~ ' Kiui. n Muxininm SWdard oiafjQn 'harm culth ur lrM~, ium 2S 57,62 ]444 3199 U«iu,mm 6.1 525 ?l,~ , 4971 b h iaity 1 5$.$ 72.71 3 rya x.801 I M&O L r r. t T ! 1 a De!tri cidth'ar r - _ t T ~ - •r T r•rr AMO21M 6 9336 Vi ~-~~~ 3 2.93384 `471542l ~ ~ 6516542 3I963 _ 11 1 1• 2 1 1 1 r 1 1 heiidty 71 .0 3.285I2 r ~ r t T B3 A.J.2. Tlh m viIut, sad r ri~Iio i (5IDr) and uric (rniD, ) of spa mIrii ftibut s o[m n £n l3 (wri and 3whori cukth r) y ~ 1ay 1? oclaniqu a assnvo1b1 CV I CoIar camera b4 i1 Y Yunw a Men Minimum MI' urA Sindard depiaOop --r-Trr ri-Tri11 - 'l rs llvDr J6197fl 74.44t 3.742k - 11.1 r11 1 ~•1 ~~ ~~ 4 D, tam 6M6 7 52,3f '4 7l.4 3.751656 1.11 r. _- -- -Tr•: _1 r• r 1 1 1• ~ ~ 11 ~.~~ Arc a~mr 11.431 X1 ~ 1 ~ 1 1 1 1 r r• r r 1 1 ~ ~ 1 ~ • e~rr c11111~~r AMD, mm 5L2914 74 42 Q7 ' 3 311 S G 'lD, mm 1 $14 59.129 709 131 2975545 r l r .r r tr. t.- t -,1 1 1 1 s 1 11 1 5I3%5 4 8 34 6354193 3 79133 Sph city 65.2003 S .QN 7] .44# 6 3.5067 Arn (mn) 7353H b, 3 7,965 1Q395,~3 1129,6 4 294 A-313+ The Ix inn v1kcs, stiadard ton (S,DI) Ruh tar (i i , r, ) of ornc g ani;tri tüibtiits of imi h] (,du ad D4uhu;hri culrivir) anal}^a by it IClirii us using kur di crut U CVS acid UY figltitiiig re inie UV wDera hard CYO parameters 14~ean Iini n~m M azlmum and d dni ChaiicuIflir l i r r• r 1 r1• 1 T AT, üuii 59r141 51.13Y(ILfl 2 ~V Y~Y IY11 I I Y LL 93U92 28 r r-~ r r. r 1. 1f 1 L _I i . (L) 2113137 492 7422 7 45 973 lha thin ii~ it Aid 50%5 11.1 .8176 4.~SX J t T• 1,R L Vll t 663359 6113615 72.05 446755 Arcs u 7,1 LI 1O 4 6I 1 L~5 A314, Th in YAN ci, 5tikrd i~Cion (I,t,) and rmigc (min, iax) of ~amc gamctric at of mango fhiit (au*a d I)Rsk tri il1ivar) 11I dxl by If' tahniqu iiag Nfk CV I _ . NIR urn b d Y Piriind R [can MiM'mam fnxlmurn Blanch dcrlitum 'Itiaur~ c~]Lirar 1.1I 57.24 •9 t 1 4143 64111,E 52.776 ] .4 7~ 4.O64~133 1 14N (JiLLL') 52 r. 4366 4l 646.N25 MAO cold r 67. 1$ .112 74.571$1 3. I12 •GMD, inm 61I9171 56 91 6977332 3.14921 AR 57.11 49A3l 6 71i4 1 Sph~xi i~v .1 '? I rar 26 88 IN 30'~,l .9194