Application of Deep Learning Along Directional Image Gathers for High-Definition Classification of Subsurface Features
Y. Serfaty, L. Itan, R. Levy and Z. Koren
Event name: 80th EAGE Conference and Exhibition 2018
Session: Diffraction Modelling and Imaging
Publication date: 11 June 2018
Info: Extended abstract, PDF ( 553.94Kb )
Price: € 20
We present a novel method for decomposing different geometrical characteristics of imaged seismic data. The automation of procedures for enhancing interpreted/classified image data is achieved by applying principle component analysis (PCA) to directional (dip/azimuth) gathers, followed by deep learning (convolutional neural network) classification. The subsurface geometrical objects to be classified are reflectors (continuous structural surfaces) and different types of diffractors (discontinuous objects such as small-scale fractures and faults). This approach shows great promise in identifying subsurface structural features with high accuracy, low cost (no processing preparation is needed) and simple yet scalable implementation. Our preliminary results show superiority over other methods involved in geometrical transformation (e.g., Radon) and specular/diffraction weighted stacks.