基于神经网络的小波域超声信号消噪技术研究

为了提高超声无损检测（UNDT）与无损评价（UNDE）基础数据的信噪比（SNR），提出了一种基于神经网络模式识别理论的小波域超声信号消噪技术.该技术在研究材料内部散射体引起的结构噪声产生机理，以及分析传统裂谱分析(SSP)算法局限性的基础上，利用小波变换方法将原始超声检测信号分解到小波空间，并通过径向基函数（RBF）神经网络所构成的信噪分离器对信号和噪声进行识别、分离来消除噪声，得到高信噪比的超声回波信号.实验结果表明，与传统裂谱分析算法相比，该技术提高了消噪性能的稳定性，增强了湮没材料内部各类散射体散射中的缺陷回波信号能力

Study on denoising techniques for ultrasonic signals in wavelet domain based on neural networks

To enhance the signal to noise ratio (SNR) of fundamental ultrasonic echo signals for ultrasonic nondestructive testing (UNDT) and ultrasonic nondestructive evaluation (UNDE), an improved technique to suppress structural noises of ultrasonic signals was presented. After the formation mechanism of structural noises was studied and the shortcomings of classical split spectrum processing (SSP) algorithm were analyzed, the fundamental ultrasonic signals were decomposed into wavelet domain by discrete wavelet transform. A signal and noise separator based on the radial basis function (RBF) neural network was used to distinguish the target signals from the noises in wavelet domain, and the target signal was reconstructed to realize the aim of enhancing SNR by removing noises. The experiment results indicate that the presented technique has high denoising performance reliability and improves the SNR enhancing ability for ultrasonic target echo signals contaminated by structural noises compared with the classical SSP algorithm.