超声检测信号中往往会携带部分噪声信号,以材料晶界散射噪声和系统噪声居多。针对一些传统超声信号降噪方法的局限性和不足,将奇异谱分析(SSA)算法引入到超声信号的降噪中。该方法源于主成分分析法(PCA),根据奇异谱中信号主成分和噪声成分的奇异值差异提取出信号主成分,再对提取出的若干个信号主成分进行信号重构,实现降噪目的。最后对比了SSA方法与小波阈值去噪、EMD (经验模态分解)滤波和稀疏分解重构等传统降噪方法的降噪效果。试验结果表明,SSA算法对不同信噪比的含噪信号均有较好的降噪效果,显著优于其他传统的降噪方法,且无需更多的先验信息。

Some noise signals are often carried in ultrasonic detection signals, and the most of them are the scattering noise at material grain boundary and system noise. In view of the limitations or shortcomings of some traditional methods of ultrasonic signal noise reduction, this paper introduces the singular spectrum analysis (SSA) algorithm to the noise reduction of ultrasonic signals. The method originates from principal component analysis (PCA). The main component of signal was extracted according to the difference of singular value between the main component and noise component in singular spectrum, and then several extracted signal principal components were reconstructed to realize the purpose of noise reduction. The noise reduction effect of SSA algorithm is compared with traditional methods such as wavelet threshold denoising, EMD filtering and sparse decomposition reconstruction. The experimental results show that SSA algorithm has better noise reduction effect on different SNR signals, which was significantly better than other traditional noise reduction methods, and no more prior information was needed.