Deep learning pre-stacked seismic velocity inversion using Res-Unet network

Abstract

Ocean carbon storage is one of the effective ways to achieve emission peak and carbon neutrality. It requires detailed characterization of the seabed reservoir. Seismic exploration is a method of using artificially excited seismic waves to identify subsurface structures. It is widely applied in hydrocarbon exploration and geological engineering, such as reservoir prediction, structural interpretation and subsurface cavity investigation. Currently, researchers investigate the application of the method to carbon storage. Stratum velocity is the key result of seismic data processing and imaging, which determines the accuracy and resolution of the stacked profile. It ultimately affects the results of geological structure identification. This paper proposed a new deep learning velocity inversion method with the convolutional network structure and mix loss function. The results illustrated that our inversion method has better accuracy and resolution than traditional convolutional networks, and is more suitable for velocity inversion.