Seismic Facies Interpretation Based on Super-Resolution Learning

With the advancement of computer technology, deep semantic segmentation-based automatic seismic interpretation has made significant progress. Semantic segmentation methods often require high-resolution (HR) input to achieve accurate classification results. However, seismic images often suffer from low-resolution (LR) quality. Reconstructing HR seismic images from LR ones and performing accurate automated seismic interpretation are crucial for assessing subsequent oil and gas reservoirs. In this letter, we propose a multitask seismic facies interpretation network based on super-resolution (SR) learning, named SR semantic segmentation network (SRSS-Net). It consists of two subnetwork branches: the SR Reconstruction branch, which effectively reconstructs seismic images from LR inputs, and the SR semantic segmentation branch, which interprets seismic facies at an SR level. We first introduce the basic theory of SR reconstruction, and then detail the network structure and optimization strategy of SRSS-Net. For implementation, we apply SRSS-Net to the Netherlands F3 Dataset and the New Zealand Parihaka Dataset. The results indicate that it achieves favorable seismic facies interpretation results even with LR inputs.