Swin Transformer for Seismic Denoising
Fang Li, Hailong Liu, 伟 王, Jianwei Ma
Abstract
Seismic noise suppression is an important preprocessing stage for obtaining high-quality seismic signals, which are crucial for seismic exploration. Deep learning methods have achieved excellent results in the field of seismic signal processing. Currently, many researchers have used convolutional neural networks for seismic signal denoising, but few have used Transformer model for related research. We apply the swin transformer model, an improved version of transformer model based on the self-attention mechanism, to denoise two-dimensional seismic data. The swin transformer calculates self-attention within shifted windows, effectively improving information exchange within the different windows. It performs well in suppressing random seismic noise to improve the signal-to-noise ratio.