Dip-Guided Multi-Stage Transformer-Based Seismic Image Super-Resolution Network for Seismic Fault Detection under the Condition of Low Signal-to-Noise Ratios

The source file for review paper "Dip-Guided Multi-Stage Transformer-Based Seismic Image Super-Resolution Network for Seismic Fault Detection under the Condition of Low Signal-to-Noise Ratios" Abstract:Faults are key geological structures that reveal subsurface fractures and deformation characteristics. Seismic coherence attribute analysis, as an effective fault detection method, often faces limitations in low signal-to-noise ratio (SNR) regions. To address this challenge, we propose the use of super-resolution reconstruction techniques to enhance the resolution of seismic images, thereby improving fault detection accuracy in low SNR regions. Building on this, we propose a dip-guided multi-stage Transformer-based seismic image super-resolution reconstruction network (MST-SR). This method progressively enhances seismic resolution while preserving seismic wavefield characteristics and accurately recovering key geological structures such as faults and folds. MST-SR consists of three stages. The first two stages focus on feature extraction, where a U-shaped Transformer (U-Transformer) architecture is designed to replace the conventional U-shaped convolutional network (U-Net), enabling more effective capture of global features in seismic images. The third stage is responsible for feature fusion and reconstruction, designing a progressive enhancement module to optimize cross-stage feature aggregation and mitigate feature loss. Additionally, to further improve seismic resolution enhancement performance, we design a multi-scale attention module between stages, which enhances the weighting of important geological features, effectively suppresses noise interference. Furthermore, we design a fault dip-constrained loss function to ensure accurate recovery of fault geometry. Experimental results on both synthetic and field seismic data demonstrate the superior performance of MST-SR in seismic resolution enhancement. Moreover, seismic coherence attribute analysis demonstrates that the high-resolution field data obtained by MST-SR can accurately characterize fault structures even in low SNR regions.