GRU-Unet for prestack three-parameter inversion

Deep learning (DL) technology has been widely applied in geophysical inversion. In pre-stack three-parameter inversion (Vp, Vs, Den), it is common to utilize seismic angle gathers annotated with elastic parameter data obtained from well logging as inputs to learn the nonlinear mapping from seismic data to elastic parameters. A key focus of inversion work is enabling networks to learn information about low-frequency long-term dependencies and high-frequency weak signals. This paper proposes a GRU-Unet, which combines the temporal modeling capabilities of GRU networks for capturing long-term dependencies with the high-resolution spatial segmentation abilities of U-net networks, making it sensitive to seismic data characteristics. Additionally, an attention mechanism is incorporated to enhance the network’s feature information extraction capabilities.