Salt body identification using VAC-U-Net

Accurately locating subsurface salt structures is crucial for efficient hydrocarbon exploration and production. However, conventional deep learning methods still struggle with accurately delineating salt boundaries and preserving detailed structural features. This paper proposes VAC-U-Net, an improved U-Net architecture for enhanced salt identification. This model uses the first 13 convolutional layers of the VGG16 network as an encoder to extract image features, and incorporates an atrous spatial pyramid pooling (ASPP) module with residual connections to enhance the capture of multi-scale contextual information. A content-guided attention fusion (CGAFusion) module, incorporating channel, spatial, and pixel attention mechanisms, is then introduced to effectively integrate multi-level information from key regions and boundaries and thereby enhance the interaction between high-level and low-level semantic information. Salt segmentation is ultimately achieved using a multi-level upsampling and decoding structure. TGS data validation achieves an intersection over union of 85.49%, pixel accuracy of 96.21%, and F1-score of 91.84%. Compared with its original counterpart, our model shows significant improvements in pixel accuracy and boundary restoration, demonstrating better robustness and generalization capability. It provides effective technical support for subsurface salt identification.