基于MEA-BP神经网络对木材内部缺陷诊断的研究  

Study on the Diagnosis of Internal Defects of Wood Based on MEA-BP Neural Network

作　　者：刘佳美 徐凯宏[1] 王立海 

机构地区：[1]东北林业大学机电工程学院 [2]College of Engineering and Technology, Northeast Forestry University

出　　处：《林产工业》2018年第2期19-24,共6页China Forest Products Industry

基　　金：国家林业局948项目“便携式立木腐朽电阻断层成像关键技术引进”(41314201)

摘　　要：为了提高木材内部缺陷的自动识别率,采用电阻层析成像(ERT)的方法获取电导率波动信号,通过小波包变换对采集的数据进行3层小波包分析,对八维特征向量进行提取,利用思维进化算法(MEA)优化权值和阈值,孔洞、节子、腐朽试样各45组数据,进行BP神经网络训练,每种缺陷20组作为测试集,识别木材内部缺陷。结果表明:MEA-BP神经网络对木材孔洞、节子和腐朽的识别率分别为96.92%、95.38%和92.31%,该模型解决了复杂组合的优化问题,提高了搜索效率,并且达到最佳的预测效果。In order to improve the automatic recognition rate of wood internal defects, Electrical Resistance Tomography （ERT） method was used to obtain the electrical conductivity fluctuation signal. Three-layer wavelet packet analysis is performed on the collected data by wavelet packet transform, and the 8 dimensional feature vector was extracted. The weight and threshold were optimized by using Mind Evolutionary Algorithm （MEA）. Hole, knot and decay of the 45 groups of data for BP neural network training, 20 sets for each defect was used as a test set, and the defects of wood were identified. The results showed that the recognition rates of MEABP neural network for wood holes, knots and decay were 96.92%, 95.38% and 92.31%. The model solves the optimization problem of complex combination, improves the search efficiency and achieves the best prediction effect.

关 键 词：缺陷识别 小波包分析 MEA-BP神经网络 无损检测 

分 类 号：S781.5[农业科学—木材科学与技术]