Learning velocity model for complex media with deep convolutional neural networks

The problem of identifying elastic media properties based on their measured response is a well-known one. This problem has many applications and variations in industrial non-destructive testing, seismic exploration, biomedical engineering, and other areas. This paper considers methods based on acoustic or elastic wave excitation in a media under consideration, recording the media's response and identifying the media's properties from this response. This problem statement is typical for ultrasonic techniques and seismic imaging. There are many different approaches for solving an inverse problem to determine the spatial distribution of mechanical properties from the recorded response. New methods have emerged recently based on the success in deep convolutional neural networks research and development. The media's response is used as an input for the