Abstract:

A novel method of softwood species computer automatic recognition through cross-sectional microscopic images is proposed in this paper. The method extracts PCA(principle component analysis)feature of wood images, generate "EigenTrees", and then use SVM(support vector machine)to classify samples in feature space. Eight kinds of softwoods species, twelve samples in each species are used in our experiment. Using leave-one-out cross-validation(LOOCV), wood recognition experiments are carried out under different conditions on image split methods, classification algorithms of nearest neighbor and SVM, and various norm distances. The results of these experiments show that wood recognition by parts of wood micro-texture is possible under certain conditions.

Key words: principle component analysis, support vector machine, computer vision, softwood recognition, EigenTrees