2020 Volume 66 Issue 3 Pages 171-182
Near-infrared spectroscopy enables identification of previously inseparable wood species by the observation of their anatomical features by optical microscopy and through the detection of subtle differences of chemical components between wood species. This research verified whether the combination of Near-Infrared spectroscopy, multivariate analysis and feature selection method can discriminate the anatomically similar and important softwood species used as building materials, Chamaecyparis obtusa and Thujopsis spp., and Tsuga sieboldii and Tsuga heterophylla. In addition, classification of old C. obtusa and T. dolabrata specimens collected from traditional buildings was conducted based on the statistical model trained by their new samples. In the case of new specimens, classification accuracy reached almost 90% in Cupressaceae and Tsuga, respectively. Moreover, feature selection succeeded in the quantitative demonstration of important wavelength regions for their discrimination. On the other hand, classification accuracy of old specimens had deteriorated to some degree because of the transition of chemical components during their degradation. However, relatively high classification accuracy was kept because the important variable regions chosen by the model are robust to the aging process by chance.