基于支持向量机的超声检测缺陷识别研究

摘要

In order to solve the problems of poor generalization ability and over learning that the traditional artificial neural network suffers in flaw recognition in ultrasonic testing field when the sample sets are small, a new flaw recognition method based on support vector machine is proposed. Wavelet decomposition is applied to de-noising of the signals. Then wavelet packet transform is utilized in feature extractio. At last a multi-class support vector machine is constructed to identify flaws. The experimental result suggests that the support vector machine method has advantages of high accuracy ratio of flaw recognition and strong generalization ability and it can be applied in flaw recognition in ultrasonic testing.%为解决超声检测领域传统人工神经网络方法对于小样本进行缺陷识别时存在的泛化能力差和过学习等问题,提出了一种基于支持向量机的超声检测缺陷识别方法.先使用小波分解对信号进行降噪,再使用小波包变换提取特征值,构造多类分类支持向量机进行缺陷识别.实验结果表明,支持向量机方法具有识别率高、泛化能力强等优点,能够应用于超声检测缺陷识别领域.