Classification of wood species using spectral and texture features of transverse section

Abstract

Classification of wood species using computer-related technology is not only significant in the modernization of wood industrialization, but also makes it easier for nonwood professionals to identify different wood types accurately and avoid being deceived. Presently, most wood classification methods use a single feature to describe the wood, which is restrictive and provides incomplete information. In this study, macroscopic transverse section images and spectral curves of wood were used as the research objects, and 50 species of wood were classified. A wood species identification method based on the fusion of spectral features and texture features is proposed, which has the advantages of convenient data collection, fast identification speed, high identification accuracy, and anti-noise interference. First, a digital camera and a spectrometer were used to acquire the image and spectral curve from the wood transverse sections. The acquired spectral and texture features of the wood transverse sections were extracted with the fractal method and the local binary pattern theory method, respectively, and both extracted features were fused using the canonical correlation analysis feature fusion method. The fused features were then classified using a support vector machine classifier. The experimental results demonstrated that the classification accuracy of the texture and spectral features alone was 91.96% and 92.67%, respectively, whereas that of the fused features was 99.16% in the “leave-one-out” cross-validation. The wood classification method outlined in this paper has higher classification accuracy than existing mainstream methods. In addition, even after adding noise to the image and spectrum, it was observed that the classification accuracy did not decrease significantly, which indicates that the method described in this paper achieves excellent classification even in the presence of noise interference.