欢迎访问林业科学,今天是

林业科学 ›› 2006, Vol. 42 ›› Issue (8): 63-68.doi: 10.11707/j.1001-7488.20060811

• 论文及研究报告 • 上一篇    下一篇

基于小波神经网络的木材内部缺陷类型识别的研究

齐巍 王立海   

  1. 东北林业大学,哈尔滨150040
  • 收稿日期:2005-07-25 修回日期:1900-01-01 出版日期:2006-08-25 发布日期:2006-08-25

Identifying the Patterns of Defects in Timber Using Ultrasonic Test Based on Wavelet Neural Networks

Qi Wei,Wang Lihai   

  1. Northeast Forestry University Harbin 150040
  • Received:2005-07-25 Revised:1900-01-01 Online:2006-08-25 Published:2006-08-25

摘要:

利用榆木标准试件,在实验室内用超声波检测仪器对试件进行缺陷分类检测,检测信号作为原始信息。各类试件的原始信号用小波包分解,计算缺陷试件与完好试件在小波包第5层各结点的信号能量变化值。试验发现:木材缺陷引起能量的变化值主要由木材缺陷的大小或严重程度来决定,亦即木材的缺陷程度越严重,能量的变化幅度就越大;对小波包5层分解后各信号结点的能量变化值进行分析,发现在32个结点中,(5,0)结点在各类缺陷试件中能量值变化最大;使用经小波压缩后的信号作为神经网络的输入,形成应用频带能量变化值和应用(5,0)结点小波包系数的2个不同输入特征的人工神经网络。对比分析2个网络识别木材缺陷类型的能力,(5,0)结点小波包系数作为特征训练得到的神经网络检测精度更高。

关键词: 木材缺陷, 超声检测, 小波分析, 人工神经网络

Abstract:

Nondestructive testing(NDT) for wood inner-defect detecting, combined with wood sciences, electronics, signal procurement and processing, and pattern diagnosing, are very important for timber production, wood processing, evaluation of standing trees and assessment of wooden structures. This paper carried out the indoor experiments for NDT of Elm wooden test samples using ultrasonic instrument in order to identify the inner-defect patterns. Wavelet transform and wavelet packet analysis was employed to identify the characteristic values of defect signals. The original signals were decomposed, and then the energy varieties of traveling signals for different layers were calculated. The energy spectrum variety of ultrasonic signals at layer 5 were taken as the eigenvalues of transform matrix. The results of test showed that :1) The energy spectrum changes of a ultrasonic signal is proportional to the degree of defects in wood; 2) Energy spectrum changes at crunode 32 of layer 5 are the mostly significant compared with those at other crunodes; 3) Taking the energy varieties of signals at crunode 32 of layer 5 and the (5,0) crunode's wavelet radix as the character inputs of the artificial neural network respectively, the latter network for identifying the defect patterns works more efficiently than the former one with accuracy rate over 90%.

Key words: wood inner-defect detecting, ultrasonic testing, wavelet analysis, artificial neural network