Abstract: [Objective] In order to obtain the internal structure form of wood and improve the identification rate of internal defects in wood,according to the computed tomography image,a method of identification and classification of wood interior defects based on convolution neural network algorithm is proposed,which realizes the automatic classification of wood efficiently.[Method] First of all, the internal cross-sectional CT images of wood samples on independent design were obtained, and the computed tomography system equipment was developed. Then,700 original sample images are randomly select after processed these sample images,from which 20 000 sample images of single defect region can be intercepted. The data set is expanded to 70 000 images by image enhancement algorithm. The size of normalized image is 28×28 pixel,and it can be divided into four parts:normal,cracked,insect hole and knot,from which 60 000 pictures of them are taken as training set,and the remaining 10 000 pictures are taken as test set. The remaining 100 images are used to implement doing experimental test.[Result] 60 000 sample images are used to train the network model,and the 10 000 sample images are classified. The classification accuracy is up to 99.3%. The accuracy of the average classification is 95.87% by verifying the remaining 100 original sample data.[Conclusion] The classification method that based on convolution neural network algorithm overcomes the problem of complicated image preprocessing,complex training method,numerous training parameters and amounts of time consuming and so on. It has the advantages of high precision,low complexity and good robustness. The identification accuracy and time are more accurate and shorter than those of the current conventional algorithms. It is a non-destructive,efficient and accurate classification method.

Key words: wood, nondestructive testing, convolution neural network, image identification