Wavelet-enhanced TEM1Dformer denoising network to reduce noise in TEM signals

Abstract

Transient electromagnetic method is widely used in the field of geophysical exploration. But the interference of noise poses a challenge to the accurate analysis and application of TEM signals, so it is necessary to denoise the signal. However, the signal processing capability of the existing EMD-like and VMD-like methods traditional methods is insufficient. In addition, the smoothness constraints of denoising results of signals processed only by the deep learning method is poor, and it cannot be effectively expressed on field signals. To solve these problems, this paper proposes a Wavelet-Enhanced TEM1Dformer Denoising Network (WE-TEM1Dformer) to improve the smoothness constraints of signal processing and signal adaptability. The wavelet thresholding algorithm is a preprocessing step to model the global correlation of signal features using the Transformer module that introduces a local attention mechanism. After comparison and verification, this method enhances the processing capability of non-smooth features, improves the accuracy and robustness of TEM field signal denoising. The experimental validation is carried out in the field of an iron ore geological exploration area in the central region of China, and the results show that the data interpretation accuracy of the WE-TEM1Dformer network is effectively improved, and the validity and accuracy of the present model are better verified.