超声检测缺陷分类中信号处理方法的探讨

根据检波信号与射频信号所含信息量的不同，针对超声检测缺陷回波中的这两种信号进行了实验分析：用典型的金属缺陷信号来做比较，把同时采集到的两类信号分别进行特征提取，并用“类别可分性判剧”做定量对比；分别用BP网络和RBF网络对检波信号提取的特征值为依据进行缺陷分类，来比较这两种网络的性能差异。最后实验表明；基于射频和检测输出的缺陷信号，其可分性指标之间的差别并不明显；RBF网络比BP网络具有更快的学习速度，同时能够有效的提高分类器的泛化分类准确率。

Comparing Study on BPN with RBFN and Demodulated Signal with Radio-frequency Signal for Flaw Classification in Ultrasonic Testing

According to difference in information between demodulated signal and radio frequency signal, those two kinds of echo signal in ultrasonic testing were analysed through experiments. Using some typical metal flaws, characteristic extraction of the two kinds of signals were acquired at the same time, which were compared in quantity by sort separability criterion; and flaw classification was carried out based on the character values of demodulated signals by BP neural network and RBF neural network to compare the performance of these two kinds of network. Results of the experiment show that the difference between separability targets of flaw signals based on demodulated signal and radio frequency signal is not obvious; RBF network is faster than BP network in learning, and it can efficiently improve generalization classification accuracy of classifier.