Post-stack Impedance Inversion Based on CoT-UNet

Summary Convolutional Neural Network (CNN), known for their strong ability to extract local features, are widely used in seismic inversion. U-Net, a popular variant of CNN, has also achieved remarkable success. However, the inherent limitations of convolution operations make CNN less effective at capturing global and remote semantic interactions. In contrast, Transformer, which is based on the self-attention mechanism, excel at learning global relationships but often struggle to preserve local features. CoT-UNet is a hybrid architecture that intelligently combines the advantages of both CNN and Transformer, effectively overcoming the limitations of CNN while minimizing feature loss. This approach has already been validated in the field of medical image recognition. In this paper, CoT-UNet is applied to post-stack impedance inversion and compared with the original U-Net to demonstrate its superior capability.