Approaches to Recognize Cavity Inclusions in Elastic Media in Problem of Monitoring Test Sites

In the paper approach to reconstruct velocity model of elastic medium in the problem of monitoring cavity zones of underground nuclear explosions is considered. Main problem is to find location and to define characteristic size and shape of cavity object in isotropic elastic media. Cavity is presented by oval shape object formed in the result of an underground nuclear explosion. To recognize cavity on seismic field snapshots and to reconstruct velocity model geometry neural network was used. Neural network was trained with set of 2D results of full seismic field simulation for different models. We performed tests on developed algorithm in area of 3D finite difference simulation with hollow cavity inclusion described by elastic parameters with zero values. The advantages and results of using neural network approach are described.