Three-Dimensional Seismic Fault Identification Method Based on an Improved U-Net Model

Seismic fault identification is a core foundation of seismic data research and is crucial for oil and gas exploration. Traditional methods rely on seismic attribute analysis, which are inefficient and sensitive to noise. In recent years, deep learning-based fault identification methods have made significant breakthroughs, but when applied to complex seismic data, the identified faults remain coarse and unclear. To address this issue, this paper proposes an improved Multi-Scale Dilated Attention Dual-Path U-Net (MDADU-Net) model. The model utilizes multiscale convolutional extraction unit to identify fault features of different scales and incorporates dilated attention unit to capture both global information and local details. Additionally, a spatial aggregation gate is integrated into the skip connections, further enhancing the model's adaptability to real seismic data. Experimental results show that the accuracy of the proposed model on synthetic data validation sets improves by 5.28 percentage points compared to the baseline model. Testing on the Netherlands F3 block further validates the generalization capability of MDADU-Net under complex conditions, providing more precise fault identification results for oil and gas exploration.