Computational seismic interpretation using geometric representation and tensor-based texture analysis
MetadataShow full item record
In hydrocarbon exploration, seismic interpretation is a process that infers geological information from processed seismic data. With the dramatically growing sizes of collected seismic data, manual interpretation has become more time consuming and more labor intensive. To improve efficiency and effectiveness, we design a computational interpretation framework that interactively delineates important geological structures such as faults and salt domes in seismic volumes. This framework divides sections in seismic volumes into two groups, reference and predicted ones, where detection and tracking algorithms are applied, respectively. When delineating faults in seismic volumes with this framework, we employ the Hough transform to extract the geometric features of faults in reference sections and apply geological constraints to remove false features and improve the labeling process. Faults detected in reference sections are tracked through predicted sections to synthesize tracked faults with tracking vectors. In the interactive salt-dome interpretation, we design an attribute-based detection workflow to delineate salt-dome boundaries in reference sections. Along reference salt-dome boundaries, we construct texture tensors using tensor-based incremental learning, from which texture features are extracted for salt-dome tracking. To utilize the spatial correlations of neighboring reference sections, we introduce a novel tensor-based subspace learning method, tensor orthogonal locality discriminant projection with maximum margin criterion, TOLDP-MMC. Experimental results on real seismic data sets show that our interactive interpretation framework outperforms state-of-the-art methods in terms of accuracy and efficiency on fault and salt-dome interpretation.