VMB-Net: a deep learning network for velocity model building in a cross-well acquisition geometry

We propose a novel artificial neural network called VMB-Net, which is capable of estimating P-wave velocities directly from prestack common-source gathers in a cross-well acquisition geometry. The network is composed of a fully connected unit and a fully convolutional unit. The parameters in the network are tuned through supervised learning to map multi-shot common-source gathers to velocity models. To enhance the generalization ability, the network is trained on a massive data set, in which, the velocity models are modified from natural images which are collected from an online repository, and multi-shot seismic traces are simulated from those models. Shot gathers from different source positions are transformed as channels in the network to increase data redundancy. The training process is expensive, but it only occurs once up front. The cost for predicting velocity models is negligible once the training is complete. Using the trained network to predict velocity models from testing set shows encouraging results.