关键词: 材种识别, 机器学习, AlexNet, SVM

Abstract:

Objective: Based on the cell characteristics of the processed image of cell on wood end faces, the search for corresponding machine learning methods can significantly increase the accuracy of identifying and realize efficient utilization and processing of wood, with the aim to provide an important basis for the identification and protection of rare wood species. Method: Using the cells on end faces of five kinds of wood(Abies nephrolepis, Larix olgensis, Picea jezoensis var.microsperma, Liriodendron chinense, Magnolia officinalis subsp. biloba) as research samples, a variety of differential images are extracted as data sets. Feature information is extracted by image processing, and support vector machine classifier(SVM) and AlexNet neural network are used for classification and recognition. According to the difference of wood end face cell distinction, BN (batch normalization) algorithm is added to the AlexNet neural network architecture for optimization, and a more efficient wood recognition method is designed. Result: The 29 680 enhanced images are divided into a 7∶3 ratio and saved in the training set and the test set folders. The test samples are labeled and put into the same folder, then conduct an overall batch test of the three classification algorithms. The results show that the overall recognition accuracy of the test set of the support vector machine classifier is 84.67%, the overall recognition accuracy of the test set of the AlexNet neural network is 88.76%, the overall recognition accuracy of the AlexNet neural network based on the BN algorithm is 91.15%. It can be seen that the recognition accuracy of the AlexNet neural network based on the BN algorithm is better. Conclusion: When the sample size is sufficient, the classification effect of the AlexNet neural network on images of cell on wood end faces is significantly better than that of the SVM classifier. The optimized AlexNet neural network based on the BN algorithm architecture is more sensitive to the linear features of the image, retains the fitting optimization of the AlexNet neural network and speeds up the optimization rate, which can effectively improve the classification accuracy and achieve high-precision wood classification.

Key words: wood species identification, machine learning, AlexNet, SVM