几种超声NDE信号降噪方法的性能比较

Performance comparison of several ultrasound NDE de-noising techniques

为提高超声无损检测的准确性,需要对超声NDE信号中因随机分布于媒质中的大量散射微粒所引起的结构噪声进行降噪。由于信号和噪声的频谱范围基本重叠,传统的线性滤波方法不能提供理想的降噪结果。介绍了几种对超声NDE信号进行降噪的新方法:Wigner-Ville分布法、小波变换法和基于时间延迟的神经网络法,并从信噪比(SNR)、检测概率(PD)和估测深度(ED)等三个重要参数对它们的降噪性能进行计算机仿真实验的比较。结果表明:小波变换法和神经网络法的降噪效果较Wigner-Ville分布法要好。对实际信号的测试还表明,小波变换由于不像神经网络那样需要训练,是一种更为理想的超声NDE信号降噪方法。

In order to increase accuracy of the ultrasound non-destructive evaluation （NDE）, it is required to reduce structural noise caused by randomly distributed scattering-grains in the material. Since spectra of the signal and noise overlap, traditional linear filtering methods cannot produce desirable de-noising results. In this paper, three filtering methods, i.e., Wigner-Ville distribution, discrete wavelet transform and non-linear time-delay feed-forward dynamic neural network are studied. Three parameters, signal-to-noise ratio （SNR）, probability of detection （PD） and estimated depth （ED） are calculated to compare the algorithm performance in the simulation studies. It is shown that wavelet transform and neural network perform better than Wigner-Ville distribution. Experiments also show that wavelet transform is an ideal de-noising technique for ultrasound NDE signals since it does not require a training process as used in neural networks.