The deep learning-based gravity anomaly inversion method addresses the complex challenge of deriving subsurface density variation models from surface gravity anomaly data. In order to generate various geological environments and their corresponding surface gravity anomaly datasets, three-dimensional density models considering different spatial locations and density variations are created in this paper. At the same time, the residual module and spatial attention mechanism are introduced into the U-Net architecture to improve the learning ability and inversion accuracy of complex geological structures. Experimental results demonstrate that the proposed method achieves the high-precision reconstruction of density variation models in complex anomaly environments, with a model residual error lower than 3%. Additionally, the inversion results of the density change and the gravity change in the Longshoushan fault zone show that the 2022 Menyuan MS6.9 earthquake is in the middle of the positive and negative density changes, which verifies the applicability of the U-Net network in the field of gravity change data, highlighting the method’s value in the real-world environment.

The geological structure in the Red River fault zone (RRF) and adjacent areas is complex. Due to the lack of high-precision gravity data in the study area, it is difficult to obtain the distribution of materials within the Earth’s crust. In this study, a gravity data-fused method is proposed. The Moho depth model data are utilized to construct the gravity anomaly trend, and the mapping relation between the gravity field model data and the measured gravity data is established. Using 934 high-precision measured gravity data as control points, the bilinear interpolation method is used to calculate high-precision grid data of the RRF. Finally, the apparent density inversion method is used to obtain clear crustal density images across the RRF. The experimental results show that the fuses data not only reflect the regional anomaly trend but also maintain the local anomaly information; the root-mean-square error of the fused data is less than 5% and the correlation coefficient is greater than 90%. Through an in-depth comparative analysis of density images, it is found that the low-density anomalous zones, with depths of ~20 km in the northern and southern sections of the RRF, are shallower than those in the middle. The data-fused method provides a new way to process geophysical data more efficiently.