Recognizing the Patterns of Wood Inner Defects Based on Wavelet Neural Networks

Wood nondestructive detection technology is a new interdisciplinary technology, which has been successfully applied in wood production, wood processing, wood structure detection and many other fields. In the paper, ultrasonic nondestructive testing for wood defects is studied based on the energy spectrum variety of the ultrasonic signals by means of wavelet transform, coefficient of wavelet node and the artificial neural networks. The original signals of different elm specimen are dispelled by wavelet packet, and the signal energy variety of crunodes in the 5^th layer wavelet bundle of both defect specimen and normal specimen without any defect is obtained. The experiment results show that the energy change of defect wood specimen mostly depends on the degree of wood defects. And the defect degree is proportional to the energy change. By comparing the energy variety of every signal crunode in the 5^th layer wavelet bundle, it is explicit that the variety of the crunode (5,0) among 32 crunodes is the biggest. And the crunode contains defect character information mostly. The energy varieties of 32 crunodes in the 5^th layer and wavelet radix of (5,0) crunode are respectively regarded as the character inputs of the artificial neural networks (ANN). Two ANN networks are analyzed according to the ability of identifying wood defect patterns through training network. The identifying results show that taking wavelet radix of (5,0) crunode as the character input is more effictive in recognizing the defect patterns of wood inner defects.