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 :基于射频和检波输出的缺陷信号 ,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 According to the difference between the information contained in the detected signal and the RF signal, two kinds of signals in the ultrasonic echo defect detection are analyzed experimentally. The two kinds of signals collected simultaneously are compared respectively with the typical metal defect signal Feature extraction and quantitative comparison with “category separability judgment”. The differences of performance of the two networks were compared by using BP network and RBF network respectively to classify the defects based on the eigenvalues ​​of the detected signals. The experimental results show that the difference between the divisibility indexes based on the radio frequency and the detection output is not obvious. The RBF network has a faster learning speed than the BP network and can effectively improve the classification accuracy of the classifier rate