特征值评价对回波信号识别效果的影响

英文篇名：INFLUENCE OF FEATURE VALUE EVALUATION ON THE DISCRIMINATION OF ULTRASONIC SIGNALS
中文刊名：无损检测
英文刊名：Nondestructive Testing
作者：李萍 ; 李喜孟 ; 王来
英文作者：LI Ping ; LI Xi-meng ; WANG Lai(Department of Material Science & Engineering ; Dalian University of Technology ; Dalian Liaoning 116024 ; China)
中文关键词：超声检验 ; 信号处理 ; 特征值评价 ; 主成分分析 ; 单因素分析
英文关键词：Ultrasonic testing ; Signal processing ; Feature value evaluation ; Pricipal component analysis ; Single factor analysis
出版日期：2003-11-25
机构：大连理工大学材料工程系,大连理工大学材料工程系,大连理工大学材料工程系辽宁大连　116024 ,辽宁大连　116024 ,辽宁大连　116024
年：2003
期：11
出版单位：无损检测

摘要

应用单因素及主成分分析方法对超声回波信号的统计特征值进行选择,并以两类方法得到的最佳特征子集对实验样本进行识别。结果表明,利用特征值评价进行回波信号的识别是切实可行的。单因素分析中各个特征对回波信号的识别能力有所不同,标准偏差、最大幅值及方均根的识别率依次是67%,33%和67%。相比之下,主成分分析法得到的识别效果最好,可达97%。
Single factor analysis and principal component analysis were used to choose the statistical feature value of ultrasonic signals, and the optimum feature subset obtained by the two analysis methods was utilized to distinguish experimental samples. Results showed that it was possible to distinguish ultrasonic signals by feature value evaluation, and the distinguishability of principal component analysis was up to 97% and was much better than that of single factor analysis(standard deviation analysis was 67%, maximum value analysis was 33% and mean square root analysis was 67%).

引文

[1]　RoseJ.Flawclassificationinweldedplatesemployinamulti dimensionalfeature baseddecisionprocess[J].MaterialsEvaluation,1984,42(2):433-438.
    [2]　RoseJ.Elementsofafeature basedultrasonicinspectionsystem[J].MaterialsEvaluation,1984,42(2):210.
    [3]　刘镇清,易　勇.基于数字信号处理器的超声无损检测[J].无损检测,2001,23(12):535.
    [4]　李成林.复合材料超声检测信号的小波分解与人工神经网络识别[D].上海:同济大学,1997.
    [5]　陈彦华,李明轩.利用人工神经网络实现缺陷类型识别[J].应用声学,1998,17(2):1-4.
    [6]　KarpurP,ShankarPM,etal.Splitspetrumprocessing:optimizingtheprocessingparameterusingminimization[J].Ultrasonics,1987,25(4):204.
    [7]　李家伟,陈积懋,林维正,等.无损检测手册[M].北京:机械工业出版社,2002.168.
    [8]　王碧泉,陈祖荫.模式识别、理论、方法和应用[M].北京:地震出版社,1989.28.