Hwang and Sugiyama Plant Methods (2021) 17:47 https://doi.org/10.1186/s13007-021-00746-1 Plant Methods REVIEW Open Access Computer vision-based wood identifcation and its expansion and contribution potentials in wood science: A review Sung‑Wook Hwang1 and Junji Sugiyama1,2* Abstract The remarkable developments in computer vision and machine learning have changed the methodologies of many scientifc disciplines. They have also created a new research feld in wood science called computer vision‑based wood identifcation, which is making steady progress towards the goal of building automated wood identifcation systems to meet the needs of the wood industry and market. Nevertheless, computer vision‑based wood identifcation is still only a small area in wood science and is still unfamiliar to many wood anatomists. To familiarize wood scientists with the artifcial intelligence‑assisted wood anatomy and engineering methods, we have reviewed the published main‑ stream studies that used or developed machine learning procedures. This review could help researchers understand computer vision and machine learning techniques for wood identifcation and choose appropriate techniques or strategies for their study objectives in wood science. Keywords: Convolutional neural networks, Computer vision, Deep learning, Image recognition, Machine learning, Wood identifcation, Wood anatomy Background the feld, wood identifcation is performed by observing Every tree has clues that can help with its identifca- macroscopic characteristics such as physical features, tion. Leaves, needles, barks, fruits, fowers, and twigs are including color, fgure, and luster, as well as macroscopic important features for tree identifcation. However, most anatomical structures in cross sections, including size of these features are lost in harvested logs and processed and arrangement of vessels, axial parenchyma cells, and lumber, so anatomical features are used as clues for wood rays [3]. In the laboratory, wood identifcation is per- identifcation. Fortunately, the International Associa- formed by observing various anatomical features micro- tion of Wood Anatomists (IAWA) has published lists of scopically from thin sections cut in three orthogonal microscopic features for wood identifcation [1, 2]. Tese directions, cross, radial, and tangential [4]. Wood iden- lists are the fruits of the work of wood anatomists and are tifcation is a demanding task that requires specialized well established, so they can be used with confdence to anatomical knowledge because there are huge numbers identify wood. of tree species, as well as various patterns of inter-species Conventional wood identifcation is performed by variations and intra-species similarities. Terefore, visual visual inspection of physical and anatomical features. In inspection-based identifcation can result in misidenti- fcation by the wrong judgment of a worker. Unsurpris- ingly, this is a major problem at the forefront of industries *Correspondence: sugiyama.junji.6m@kyoto‑u.ac.jp where large quantities of wood must be identifed within 1 Graduate School of Agriculture, Kyoto University, Sakyo‑ku, a limited time. Kyoto 606‑8502, Japan Full list of author information is available at the end of the article © The Author(s) 2021. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Hwang and Sugiyama Plant Methods (2021) 17:47 Page 2 of 21 Te spread of personal computers has triggered a timbers, and screening for fraudulent species [30–33]. major turning point in wood identifcation. Wood anat- However, it is practically impossible to train a suf- omists have created a new system called computer- cient number of feld identifcation workers to meet the assisted wood identifcation [5] by computerizing the demands of the feld. Wood identifcation requires expert existing card key system [6, 7]. Several computerized key knowledge of wood anatomy and long experience, so databases and programs have been developed to take even if a lot of money and time is spent, there are practi- advantage of the new system [8–10]. Because of the vast cal limits to the training of skilled workers [34]. biodiversity of wood, the deployed databases generally To answer the demands in the feld, various approaches cover only those species that are native to a country or a have been proposed, such as mass spectrometry [35–37], specifc climatic zone [8, 9, 11, 12]. Although this system near-infrared spectroscopy [38, 39], stable isotopes [40, has made the identifcation of uncommon woods easier, 41], and DNA-based methods [42, 43]. However, these traditional visual inspection was preferred for efciency approaches have practical limitations as a tool to assist or reasons in the identifcation of commercial woods [3]. replace the visual inspection due to their relatively high Te computer-aided wood identifcation systems used cost and procedural complexity. Tis is where CV-based explicit programming, which required the user to pro- identifcation techniques and ML models can be very gram all the ways in which the software can work. Tat important. Clearly, automated wood identifcation sys- is, the user had to teach the software all the identifcation tems are urgently needed and CV-based wood identifca- rules, which was never an efcient way because there are tion has emerged as a promising system. so many rules for wood identifcation. Tis programmatic In this review, we provide an overview of CV-based nature made it difcult to spread the system globally. wood identifcation from studies reported to date. CV Over time, computer-aided wood identifcation set- techniques used in other contexts, such as wood grading, tled with web-based references such as ‘Inside Wood’ at quality evaluation, and defect detection, are outside the North Carolina State University [13] and ‘Microscopic scope of this review. Tis review covers CV-based identi- identifcation of Japanese woods’ at Forestry and Forest fcation procedures, provides key fndings from each pro- Products Research Institute (FFPRI), Japan [14]. Tese cess, and introduces the emerging interests in CV-based are very useful open wood identifcation systems that wood anatomy. cover a wide variety of woods, but require expert knowl- edge of wood anatomy. As such, there are various obsta- Workfow of CV‑based wood identifcation systems cles to the further development of computer-aided wood Image recognition or classifcation is a major domain identifcation, so this is where machine learning (ML) in AI and is generally based on supervised learning. comes in. Supervised learning is a ML technique that uses a pair of ML is a type of artifcial intelligence (AI) where a sys- images and its label as input data [44]. Tat is, the clas- tem can learn and decide exactly what to do from input sifcation model learns labeled images to determine clas- data alone using predesigned algorithms that do not sifcation rules, and then classifes the query data based require explicit instructions from a human [15, 16]. In a on the rules. Conversely, in unsupervised learning the well-designed ML model, users no longer have to teach model itself discovers unknown information by learning the model the rules for identifying wood, and even wood unlabeled data [45]. Classifcation is generally a task of anatomists are not required to fnd wood features that supervised learning and clustering is generally a task of are important for identifcation. Computer vision (CV) is unsupervised learning. a computer-based system that detects information from CV-based wood identifcation systems follow the gen- images and extracts features that are considered impor- eral workfow presented in Fig. 1. Image classifcation is tant [17–19]. Automated wood identifcation that com- divided methodologically into conventional ML and deep bines CV and ML is called computer vision-based wood learning (DL), both of which are forms of AI. In conven- identifcation [20, 21]. AI systems based on CV and ML tional ML, feature extraction, the process of extracting are making great strides in general image classifcation important features from images (also called feature engi- [22–25]. Te same is true for wood identifcation and neering), and classifcation, the process of learning the related studies have been increasing [20, 26–29]. extracted features and classifying query images, are per- Wood identifcation is a major concern for tropical formed independently. First, all the images in a dataset countries with abundant forest resources, so there is a are preprocessed using various image processing tech- high demand for novel wood identifcation systems to niques to convert them into a form that can be used by a address the wide biodiversity. Tere are various on-site particular algorithm to extract features. Ten, the dataset needs for wood identifcation, such as preserving endan- is separated into training and test sets, and the features gered species, regulating the trade of illegally harvested are extracted from the training set images using feature Hwang and Sugiyama Plant Methods (2021) 17:47 Page 3 of 21 Fig. 1 General workfow of conventional machine learning and deep learning models for image classifcation extraction algorithms.