K-SVD和OMP算法在超声信号去噪中的应用

针对在线采集时超声波检测信号中存在大量噪声,降低了材料内部缺陷诊断准确性的问题,提出了一种基于广义K+奇异值分解算法(K-SVD)和正交匹配追踪算法(OMP)相结合的超声回波信号去噪算法。该算法利用K-SVD算法将Gabor字典训练成能够最有效反映信号结构特征的超完备字典,然后基于训练完成的超完备字典,用OMP算法把一定数量的字典原子进行线性组合来构成原始信号,从而实现信号的去噪。通过仿真实验将本文方法与传统的小波阈值去噪方法进行了对比研究。实验结果表明,该方法对超声回波信号的去噪效果优于小波阈值去噪方法,且噪声越大对比越明显,不仅可更有效地滤除信号中的高斯白噪声,提高信噪比,且尽可能保留了原始信号有用信息。

Application of K-SVD and OMP algorithm on ultrasonic signal denoising

A lot of noise exists in the on-site collected ultrasonic signals, which decreases the diagnostic accuracy of internal defects in materials. For solving this problem, a kind of ultrasonic echo signal denoising algorithm which based on the combination of the generalized K-singular value decomposition algorithm (K-SVD algorithm) and the orthogonal matching pursuit (OMP) algorithm was raised. In this method, using K-SVD algorithm,the Gabor dictionary was trained to become the ultra-complete dictionary, which can effectively reflect the signal structure features. Then based on the trained ultra-complete dictionary, a certain number of dictionary atoms were combined linearly to form the original signal by OMP algorithm, and eliminating the noise finally. Through simulation, the proposed method was compared with the traditional wavelet threshold method. The results indicate that this method has better ultrasonic echo signal denoising effect than thewavelet threshold method, and the greater the noise, the more obvious the contrast.Furthermore, this method not only more effectively filters the Gaussian white noise in signals, but also improves signal to noise ratio, and retains the useful information in the original signal.