A deep learning neural network to extract of P- and S-wave transit times from Vertical Seismic Profile (VSP)

<p>This work was motivated by our need to repetitively determine thousands of transit times from ensembles of forward modelled CSG’s for use in a minimization inversion that focusses on obtaining in situ anisotropic elastic constants. By leveraging the capabilities of a deep learning neural network, specifically a three-layer U-net architecture with PyTorch, this could significantly reduce the labor intensity for similar inversion problem with good accuracy. Note: the current implementation has been trained exclusively on a specific complex 3D geological model with two contrasting rock types in a noise-free environment.</p>