Experience of Using Neural Network Neocognitron in Solving the Inverse Problem of Geophysics

The paper considered the possibility of using the neocognitron neural network for the recognition of geological physical models of the environment (GPME). An algorithm for choosing a filter for the method for finding boundaries on the brightness difference and a modified neural network model for recognizing GPME of various shapes for modifying a neural network are proposed. The developed software is demonstrated. A series of experiments to restore the structure of the GPME was carried out using this software. Recommendations are given to improve the performance of the neural network. This work was supported by the RFBR grant No. 19-07-00170.