超声检测缺陷分类的降噪及特征提取问题研究

针对超声检测回波信号中可能具有噪声干扰并难以剔除的问题，提出了利用“小波降噪”对超声信号进行处理的算法和应用“类别可分性判据”评价特征值的方法，并通过实验进行了验证。首先将小波变换用于超 声信号噪声处理，然后利用类别可分性判据对缺陷信号的特征选择进行评价，最后通过RBF网络对获得的超声检波信号进行缺陷分类以验证这种方法的有效性，实验结果表明：小波降噪算法充分利用了超声回波信号的时域、频域信息，不仅降噪效果明显，而且缺陷定位准确；类别可分性判据对缺陷信号的特征提取也起了定量衡量尺度的作用。

Study on Wavelet Denoise and Characteristic Extraction of Echo Signal to Flaw Classification in Ultrasonic Testing

There are usually two problems in flaw classification in ultrasonic testing. One is the noise in echo signal in ultrasonic testing, which is sometimes very difficult to eliminate. The other is the criterion problem for validity evaluation in echo signal characteristic extraction. A noise eliminating method for ultrasonic signal with wavelet denoise was presented and sort separability criterion was used to solve the second problem in this paper. And these two methods were verified by some experiments. Firstly wavelet transform was used in denoising process of ultrasonic signal; Then sort separability criterion was used to evaluate the characteristic choice of flaw signals; And finally the characteristic values of flaws in demodulated signal were classified by RBF neural network to validate the above methods. Results of the experiments show that due to making the best use of the information of time and frequency domain at the same time in ultrasonic echo signals, wavelet deniose algorithm not only decreases the noises obviously, but also locates flaws accuratelly. And the sort separability criterion can also play the role of being a quantification measure on the characteristic extracting of flaw signals.