AMT denoising via double sparse dictionary learning

Abstract Audio magnetotellurics (AMT) is commonly employed in the exploration of resources. However, the energy of AMT signals is relatively weak, making them susceptible to various artificial sources of interference, thus being submerged in intense noise. This poses a significant disadvantage for the identification of various resources. Sparse representation is a commonly used method for AMT denoising, but under strong noise interference, conventional sparse representation methods often fail to effectively remove noise due to the inadequate performance of the sparse basis. To enhance the performance of the sparse basis and achieve more accurate AMT denoising results, we propose a double sparse dictionary (DSD), which is formed by the cascading of wavelet transform and data-driven tight frame (DDTF) dictionaries. The wavelet transform compensates for the drawbacks introduced by DDTF, such as the introduction of new noise, while DDTF addresses the challenge of adaptive sparse representation of noise encountered by wavelet transform. Experimental results demonstrate that our proposed method achieves more precise denoising results compared to conventional sparse representation methods.