OPMNet: Overlapped Patch Merging Convolution Net for Seismic Image Super-Resolution Reconstruction

This paper introduces OPMNet, a novel deep learning framework for simultaneous seismic super-resolution and noise attenuation. The architecture features two key innovations: (1) an Overlapped Patch Merging module that maintains high-frequency details during downsampling, and (2) a hybrid MS-SSIM + L1 loss function that enhances structural preservation. Benchmark tests show OPMNet achieves 9.50% higher PSNR and 4.43% better SSIM than conventional U-Nets. Field data applications demonstrate significant improvements in fault visibility, thin-layer resolution, and spectral content. The framework effectively addresses the dual challenges of resolution enhancement and noise suppression in seismic processing workflows.