A Novel Denoising Method of Multiscale Feature Decoupled Collaborative for Marine MT Signals

Abstract Marine magnetotellurics (MT) is a significant geophysical method for probing deep seafloor structures. The dispersive attenuation characteristics of the natural geomagnetic field in dynamic marine environment gives rise to weak signals, which coupled with various noise components, significantly affects the interpretation of the data. To address the issue of composite noise suppression in marine MT signals, this study proposes a denoising method of Multiscale Feature Decoupled Collaborative (Dn‐MFDC). Initially, a two‐branch deep hierarchical convolutional architecture is constructed to handle multi‐source composite noise, optimizing the multi‐scale feature representation of marine MT signals via nonlinear mapping. Subsequently, leveraging the multi‐scale feature representation of the signal, higher order statistical properties are employed to decouple the noise from the effective signal, achieving statistical independence between them. Finally, experiments are conducted on both synthesized and field marine MT signals. The proposed method effectively mitigates composite noise across varying intensity levels and exhibits superior performance in field data experiments. The results validate the effectiveness and robustness of the proposed method, presenting a novel approach to suppressing composite noise in marine MT signals.