3D scattering wavelet transform CNN for seismic fault detection

The detection of faults is of great significance for oil and gas reservoir prediction and migration research. Deep learning methods has been gradually being used in seismic fault detection in recent years. Three dimensional (3D) scattering wavelet transform is multiscale, rotation and translation invariant, and can highlight the spatial structure of faults at different scales and directions. In this paper, we propose a 3D scattering wavelet transform CNN (3D-SWnet), which mainly contains two part. One is the 3D scattering wavelet transform, which can adequately capture the spatial information of the faults with different scales and angles, the other is 3D-Unet convolutional neural network for faults classification. This method takes full advantage of features of 3D scattering wavelet transform and the classification ability of deep learning. Based on the synthetic seismic training data, the fault detection results of other synthetic seismic data and 3D field data show the proposed method's high accuracy, clean background, and sharp fault edges.