Applications of Deep Neural Networks for Velocity Model Building

This technical report discusses the application of deep neural networks (DNNs) in the context of velocity model building (VMB) within seismic processing. It outlines the limitations of traditional methods, such as reflection tomography and full waveform inversion (FWI), particularly in complex geological settings. The authors introduce a novel approach utilizing a deep neural network architecture that integrates Fourier neural operators (FNOs) and convolutional neural networks (CNNs) to improve the accuracy of velocity updates. The methodology allows for the direct mapping of velocity errors from depth-migrated gathers, bypassing the need for prior information like residual move-out picks. The report details the training process, which incorporates a diverse range of synthetic data to enhance model robustness. Several field data examples are presented, demonstrating the effectiveness of the DNN approach in producing structurally consistent velocity models and improving data alignment. The findings suggest that this method can significantly accelerate the VMB process while maintaining geological accuracy.