Subsurface Image Morphing Operator Using Deep Learning Techniques

Abstract

Abstract Velocity uncertainty is one of the major challenges for subsurface imaging in oil & gas exploration. A surrogate migration engine based on image morphing operation can significantly reduce migration costs and speed up subsurface velocity model building workflow. We develop a machine learning based approach to predict subsurface image change due to velocity perturbation. This fast image change estimator takes the three channel inputs: an initial velocity model, its migrated image, and a velocity perturbation. It outputs the new image due to the velocity change. It is implemented using deep neural networks with 3D Fourier neural operator. We verify this image morphing operator with both synthetic data and field data experiments. The goal of this study is in speeding up the velocity model scenario tests.