Denoising method of microwave earth exploration test system based on Deep-KSVD

Abstract In the process of data acquisition of complex underground media in the microwave ground exploration test system, electromagnetic echoes are susceptible to various noises, resulting in the ambiguity of the in-phase axis of the diffraction hyperbolic curve, which seriously affects the subsequent interpretation and processing. Traditional denoising algorithms are difficult to meet the requirements of denoising complex data. To solve this problem, a microwave profile denoising algorithm based on Deep-KSVD is proposed in this paper. By combining the K-SVD sparse denoising algorithm with the idea of deep learning, the problem of easy loss of original signal and low computational efficiency in the original denoising algorithm is overcome. Firstly, the data is divided into small data blocks, and then an equivalent learnable scheme is used to replace the OMP algorithm in the tracking stage. Finally, each data block is denoised by tracking. The weighted average of the tracked data blocks is obtained, and the accuracy of the algorithm is verified by the actual data.