Research For Wood Classification Based On Subspace And Multiple Feature Fusion

There is an important significance for forestry science and related industry to identify the tree species correctly. There are some limitations in recognizing wood species using artificial tree species identification and traditional image recognition methods.In this paper, we are going to probe the feasibility of wood recognization using pattern recognition methods and achieved the following results:(1) We mix PCA with FisherTrees, then training samples are projected into PCA and FisherTrees space respectively to form the PCA and FisherTrees features, then the two features are fused through three ways, i.e. arithmetic mean, swapping transposit-ion mean and weighting mean, fusion features are more suited to identify soft wood tree species.(2) We make a experiment on common tree species such as Tacamahaca, Picea asperata and find that if kernel principal component analysis method is used to extract the wood features, the ability of classification of adaptive boost(AdaBoost) is stronger than the support vector machine (SVM), the tree species classification efficiency of former reached90.13%, and the latter is only78.32%.(3) We choose eight kinds of softwood such as Thujastandishii, Pinustaiwanensis, Chamaecyparis obtusa, Juniperus formosana, Pinus massoniana, Pinus massoniana, Metasequoia glyptostroboides, Keteleeria cyclolopis and Cedrus deodara as our exper-iment samples, kernel principal component analysis method is used to extract the soft-wood features, linear kernel function of SVM as the AdaBoost. M2fundamental classifier, when iteration numbers arrive100times, this method can accurately distinguish above tree samples.This paper prove that it is not only feasible for the application to tree species recognition using pattern recognition techniques, but also it is possible to identify tree species what is can’t be identified by using traditional methods. Pattern recognition techniques will be a key new way of wood recognition field and it is powerful complement to the traditional wood recognition field.