Seismic Layer Segmentation Models with Channel Attention Block in Carbon Storage Study

Carbon capture technology is crucial for mitigating climate change and achieving the Paris Agreement's goal of limiting global warming below 1.5 degrees. Seismic data analysis, including facies layers, plays a vital role in assessing carbon capture and storage (CCS) potential by revealing subsurface geological structures. However, manual interpreting seismic images remains a labor-intensive, expert-driven task. To address this, automated seismic layer segmentation using deep learning techniques has gained traction, but improving segmentation accuracy remains a challenge. This work proposes a novel small attention block that leverages spatial learning paths for segmentation models like U-Net and BiSeNet. This block dynamically assigns weights to different channels within a feature map, enhancing segmentation performance on seismic layer datasets. We validate the effectiveness of our method by demonstrating improved performance on two benchmark datasets: TGS salt layer and SEG facies layer. Our findings reveal that integrating our lightweight attention mechanism can yield smoother segmentation results on seismic layers. This study is part of the ongoing effort in CCS research by focusing on improved segmentation accuracy of seismic data.