Wood Identification of Two Anatomically Similar Cupressaceae
Total Page:16
File Type:pdf, Size:1020Kb
Holzforschung 2021; 75(7): 591–602 Original article Yusuke Kita and Junji Sugiyama* Wood identification of two anatomically similar Cupressaceae species based on two-dimensional microfibril angle mapping https://doi.org/10.1515/hf-2020-0079 1 Introduction Received March 26, 2020; accepted October 14, 2020; published online November 13, 2020 The culture of wood use in Japan has long flourished because the country has an abundance of forested land. Abstract: Identifying two anatomically similar species of Knowledge of the use of different species of wood in his- Cupressaceae, Chamaecyparis obtusa and Thujopsis spp., torical sites can help us learn more about our cultural is important to better understand the culture of wood use in background. An important example of this is two species Japan. However, the conventional method, which involves of Cupressaceae, Chamaecyparis obtusa (hinoki in Japa- observing their cross-field pitting, cannot identify them in nese) and Thujopsis spp., including Thujopsis dolabrata many cases. This study solves the above problem by Thujopsis dolabrata hondae introducing an anatomical criterion based on the micro (asunaro) and var. (hinoki- fibril angle (MFA). MFA values were obtained through two- asunaro). These species have been used in structural dimensional MFA images using the uniaxial optical materials in traditional buildings such as temples because anisotropy of cellulose microfibrils. A combination of the of their excellent mechanical properties and durability preprocessed MFA images and a convolutional neural (Ido 2018). However, there is marked difference in their network (CNN) yielded an accuracy nearly of 90% in use despite the similarities between them, grounded classifying these species in cases of present and old wood in preferences that ancient populations had for using specimens. Our feature extraction and classification tech- each species (Hirai 1996). Hence, the accurate identifica- niques provide a new way for describing the anatomical tion of C. obtusa and Thujopsis spp. can help us better features of wood and identifying featureless softwoods. understand the use of wood in Japanese culture, and Using the model interpretation-related methodologies of this can in turn help the historical and architec/tural the CNN, distinct features of the two wood species were sciences. partly explained by MFA anisotropy in the S2 wall induced These species are similar not only in terms of me- by the existence of pits. chanical properties, but also in their anatomies. Conven- tionally, they have been identified by observing only the Keywords: convolutional neural network (CNN); Cupres- types of cross-field pitting (Shimaji and Itoh 1982), which is saceae; microfibril angle (MFA); wood anatomy; wood widely recognized as the most important anatomical fea- identification. tures for the identification of conifers among wood anato- mists (Richter et al. 2004). Noshiro (2011), following extensive work, concluded that the types of cross-field pitting are piceoid and cupressoid in C. obtusa, and Present address: Junji Sugiyama, Graduate School of Agriculture, cupressoid and taxodioid in Thujopsis spp. However, Kyoto University, Sakyo-ku, Kyoto 606-8502, Japan identifying them using cross-field pitting remains chal- fi *Corresponding author: Junji Sugiyama, Research Institute for lenging owing to the absence of a quantitative de nition of Sustainable Humanosphere, Kyoto University, Uji, Kyoto, 611-0011, each type of cross-field pitting. Thus, methods based on Japan; and College of Materials Science and Engineering, Nanjing unambiguous criteria should be introduced for their Forestry University, Nanjing, China, identification. E-mail: [email protected]. https://orcid.org/0000- The microfibril angle (MFA) is a reliable means of 0002-5388-4925 Yusuke Kita, Research Institute for Sustainable Humanosphere, Kyoto describing the characteristics of wood because this quan- University, Uji, Kyoto, 611-0011, Japan, titative parameter can explain a number of important fea- E-mail: [email protected] tures, such as mechanical properties (Bendtsen and Senft Open Access. © 2020 Yusuke Kita and Junji Sugiyama, published by De Gruyter. This work is licensed under the Creative Commons Attribution 4.0 International License. 592 Y. Kita and J. Sugiyama: Two-dimensional microfibril angle mapping 1986; Cave and Walker 1994; Cowdrey and Preston 1966) Table : Present and old wood specimens, and information about and shrinkage behaviors of wood (Abe and Yamamoto their database. 2006; Barber and Meylan 1964; Meylan 1972). By contrast, Scientific name State Number Sample number (KYOw) features of the MFA specific to certain species of softwoods, such as MFA value itself and their transition from early- Chamaecyparis Present , , , wood to latewood, remain undiscussed except for a few obtusa , , , , , , studies (Hori et al. 2002; Lichtenegger et al. 1999). This is , , because most of the conventional methods for MFA mea- Old , surement suffer from their low spatial resolution and Thujopsis dolabrata Present , , , vulnerability to various kinds of variations in the MFA, , , , inside and between the tracheids (Donaldson 2008). , , , However, techniques of synchrotron X-ray (Lichtenegger et Thujopsis dolabrata Present , al. 1999), polarization (Abraham and Elbaum 2012; var. hondae Mannan et al. 2016; Savić et al. 2016) and polarized Raman Thujopsis spp. Old , microscopy (Gielringer et al. 2010) have a potential to overcome these shortcomings and gives us many insights through the visualization of MFA spatial distributions in a Sustainable Humanosphere at Kyoto University (https://database. manner of a two-dimensional (2D) image. These methods rish.kyoto-u.ac.jp/cgi-bin/bmi/en/namazu.cgi). Small wood blocks in enable us to directly quantify many anatomical features, only the heartwood regions, distant from the pith, were selected for such as multi-layered and pit structures, and provide a new use as present wood specimens. The old wood specimens were basis for analyzing species-specific features. Especially, collected from the Manpuku-ji temple built in the early Edo period (mid-17th century). plant anatomies based on 2D MFA images can be more All wood blocks were cut into small pieces, approximately accessible at the laboratory scale within a reasonable 4 × 1.5 × 5 mm along the radial, tangential, and longitudinal directions, experimental time. respectively. They were embedded into 100% PEG 1500. 10 μm-thick The purpose of the present research was to develop a cross-sections of the embedded samples were cut using a sliding technique for quantifying the anatomical features of wood microtome and sealed by aqueous gum-chloral mounting media. using the MFA, and to apply it to identify the anatomically similar Japanese conifers C. obtusa and Thujopsis spp. Until 2.2 Acquisition of hyperspectral images now, this is the first attempt to apply the MFA to the identification of wood. MFA measurements were per- The sections were placed on the revolving stage of a polarized optical formed using a technique similar to the above-mentioned microscope (BX53-P, Olympus, Tokyo, Japan) equipped with a × MFA imaging with polarization. A wood identification 40 objective lens (UPLFLN40XP, NA=0.75, Olympus, Japan) and a full wave plate (U-TP530, λ = 530 nm, Olympus, Japan) for empha- system was established based on a powerful image sizing interference color transitions between MFA values. Directions of recognition technique, a convolutional neural network a polarizer and an analyzer were set to a crossed Nicol condition. The (CNN), to make the best use of 2D spatial distributions of radial and tangential walls in the earlywood zone were roughly par- the MFA. The performance of the model was evaluated allel to the direction of high and low refractive index of the plate, through metrics obtained from the results of classification generally called slow and fast axes, respectively. In our experiment, of present and old wood specimens collected from MFA imaging becomes possible via retardation values of cell walls fi under the crossed Nicol observed as interference colors. In this situ- Manpuki-ji, Uji, Kyoto. Finally, the species-speci c fea- ation, the relationship between observed intensities of interference tures of wood to discriminate between C. obtusa and Thu- colors at the specific wavelength, I(λ), and retardation values R can be jopsis spp. are discussed by examining the relationship expressed as below equation. between their anatomical features and the MFA. πR I(λ) = A2 2θ 2 sin sample sin λ (1) A means an amplitude of an incident light; θsample is an angle of 2 Materials and methods directions of a polarizer and a slow axis of a specimen; λ corresponds to a light wavelength. From the above equation, an interference color is observed as a 2.1 Sample information and preprocessing color spectrum under a normal illumination (not a laser). Therefore, we can extract retardation values from interference colors through the Present and old specimens of C. obtusa and Thujopsis spp. were pre- quantification of color spectra. For achieving this quantification, pared (Table 1). All samples in this experiment were assigned the hyperspectral images (1536 × 1024 pixels) within the visible light KYOw ID registered by the xylarium at the Research Institute for region from 461 to 602 nm at intervals of 3 nm were taken by a CCD Y. Kita and J. Sugiyama: Two-dimensional microfibril angle mapping 593 camera with a liquid crystal tunable filter, LCTF (VariSpecTM,