DisRWKV: A Multi-scale Feature Interaction Network for Removing Neighboring-shot Interference in Seismic Data

Compared to traditional seismic source acquisition, the independent simultaneous source method enhances acquisition efficiency and data quality. However, its high efficiency can result in neighboring-shot interference due to the continuous firing of multiple shots within a short time interval. Furthermore, delays in autonomous acquisition nodes single-shot Synthesis make supplementary acquisition difficult. Traditional denoising methods are less effective because of the high similarity between neighboring-shot interference and the main shot, as well as their strong reliance on empirical parameters. Deep Learning methods encounter challenges such as the difficulty of obtaining training sample pairs datasets, limitations of the convolution network's receptive field, and the slow computation speed of the Transformer. To address these issues, we have developed a deep learning-based denoising process for neighboring-shot interference. We propose the DisRWKV model, which utilizes a co-wkv bidirectional attention mechanism to perform global modeling with linear complexity, thereby reducing computational complexity. The use of Multi-Channel Fusion (MCF) enhances the integration of information across different channels and improves the ability to capture multi-scale features, thereby enhancing the extraction of contextual information for more precise removal of neighboring-shot interference. Experiments demonstrate that DisRWKV can effectively eliminate neighboring-shot interference, significantly improving the signal-to-noise ratio of seismic data, with performance surpassing other methods.