Deep learning for 3D fault detection within virtual realityvisualization seismic volumes

Ali Fathalian, Marcelo Guarido, Daniel O. Trad, Kristopher A. H. Innanen

An important key for seismic structural interpretation and reservoir characterizationis the delineating faults that are considered as seismic reflection discontinuities in conventional methods. Fault detection considers as a binary image segmentation problem of labeling a seismic image with ones on faults and zeros on non-faults using a fully supervised convolutional neural network. The network is trained by using 3D synthetic seismic images and their corresponding binary labels images. The network learns to calculate features that are important for fault detection after training with a synthetic data set. We apply this method to a migrated 3D volume from Australia. The results indicate that the neural network can predict faults from 3D seismic images. Effective visualization analysis of 3D seismic data volumes is challenging because of their large volumes and highly complex nature. 3D virtual reality (VR) visualization is a useful tool that can benefit seismic data interpretation. In this paper, the seismic information extended reality analytics (SIERA) presents a seismic data visualization in an extended reality environment. Because it is highly customizable, it provides an effective way to interact with seismic data and machine learning results.