On obtaining initial approximation for full wave inversion problem using convolutional neural network

The paper considers the problem of choosing the initial approximation when using gradient optimization methods for solving the inverse problem of restoring the distribution of velocities in a heterogeneous continuous medium. A system of acoustic equations is used to describe the behavior of the medium, and a finite-difference scheme is used to solve the direct problem. L-BFGS-B is used as a gradient optimization method. Adjoint state method is used to calculate the gradient of the error functional with respect to the medium parameters. The initial approximation for the gradient method is obtained using a convolutional neural network. The network is trained to predict the distribution of velocities in the medium from the wave response from it. The paper shows that a neural network trained on responses from simple layered structures can be successfully used to solve the inverse problem for a complex Marmousi model.