Deblending of multi-source Seismic Data Based on Dual-Element Attention Mechanism Network

The high-density blended acquisition technology effectively improves the acquisition efficiency in seismic exploration, but it requires addressing the aliasing problem of its acquired data, known as the deblending problem. Among numerous deblending methods, deep learning-based denoising has gained increasing attention and research from scholars in recent years due to its efficiency in handling complex aliased data and the advantage of not requiring manual adjustment of numerous parameters. Given the complexity of blended data and the correlation between source signals, a DeamNet network incorporating a Dual-Element Attention Mechanism is proposed to achieve blended data separation. This Dual-Element Attention Mechanism can effectively extract the features of blended data for complex multi-source blended datasets, allowing for the retention of more valid signals during deblending, facilitating subsequent imaging processing. To verify the deblending performance of the DeamNet network, numerical experiments were conducted using both synthetic and real data, and its performance was compared with DnCNN and U-Net, both of which have also been applied in the deblending field in recent years. The final comparison results confirmed the effectiveness of the DeamNet network in deblending.