Applications of Deep Neural Networks for Velocity Model Building

In this work, we utilize a deep neural network comprising a series of Fourier neural operators (FNOs), convolutional neural networks (CNNs), and fully connected neural networks (FCNs) to map velocity errors within the image domain. The network is trained to perform tomographic updates based on synthetic data, effectively replacing the conventional reflection tomographic engine. We demonstrate the performance on several field data examples from different basins worldwide, showcasing how this methodology can improve or assist conventional tomography, and help condition the model for full waveform inversion (FWI).