Application of convolutional neural networks for search and determination of physical characteristics of inhomogeneities in geological media from seismic data

With the use of convolutional neural networks, we solve inverse problems of exploration seismology to determine the spatial position and physical characteristics of geological fractures, such as the proportion of excess surface and the nature of saturation. The training and validation sets were formed using numerical modeling by the grid-characteristic method on unstructured meshes in the two-dimensional case. The continuum mechanics equations were used, while the fractures were speci ed discretely in the integration domain; this approach made it possible to obtain the most detailed patterns of wave responses.