Interactive seismic facies segmentation using multi-attribute seismic data

Automated seismic interpretation significantly enhances efficiency in exploration and production workflows. This study leverages the fine-tuning of Meta AI’s Segment Anything Model (SAM) for seismic facies segmentation, with the initial primary focus on salt bodies and karst features. SAM is pretrained on massive datasets of 11 million natural images and over 1 billion masks. SAM offers exceptional generalization capabilities across diverse natural images. As natural images have different feature signatures as compared to seismic images, we tried to calibrate feature representation of SAM to the seismic features to increase the seismic interpretation capabilities. We use multi-attribute seismic data (amplitude, coherence, and curvature) as input channels to SAM, simulating the RGB structure of natural images and enhancing geological feature differentiation and that distinguishes our approach over others deep learning models and different use cases of SAM. This study will help detect and extract the geological features from seismic data with just a prompt. This will reduce the time for interpretation and increase the efficiency of geophysical workflows.