Inclusion imaging using single-shot ultrasound and convolutional neural networks

This paper consider the problem of harder inclusion localization in an elastic media. The imaging method is based on a single-shot ultrasound with a linear array. Discontinuous Galerkin method is used for direct problem modeling and obtaining wave propagation patterns in the media. Two different architectures of convolutional neural networks are used for the inverse problem. The paper provides the numerical results for the quality of the inclusion localization depending on the neural network architecture and the shape of the heterogeneity.