Leveraging Sparse Features Learned from Natural Images for Seismic Understanding

This paper aims at adapting and utilizing the broad domain knowledge from natural images and evaluate its applicability in automation, improving attributes, and seismic processing. We propose an unsupervised learning network for seismic data analysis based on features learned from natural images. First, we learn the variances of edge and non-edge features in natural images characterized by different shapes, orientation, edges, and background. Then, we use the sparse model of learned features to study and recognize salient geological structures. We show that the proposed approach can effectively detect salient areas characterized by strong or weak edges and non-edge features within a real seismic dataset from the F3 block in the North Sea, Netherlands. The preliminary results demonstrate the potential of the proposed method in highlighting important geological structures such as salt domes and weak reflections within seismic volumes and can be effectively used for computer-aided extraction of other geologic features as well.