Inversion of gravity anomalies based on a deep neural network

Traditional linear inversion of gravity anomalies is liable to encounter local minima and suffer low computational efficiency. Given this, this paper proposed a deep learning-based inversion of gravity anomalies. Specifically, two-dimensional density models of various shapes were firstly established, and gravity anomalies were obtained through forward simulation using these models to form a dataset. Then, a deep neural network was trained using the dataset. Finally, gravity anomaly data were input into the deep neural network to directly yield inversion results. Experimental results show that the inversion method proposed in this study can determine the locations and shapes of underground anomalies quickly and accurately, with high generalization ability and anti-noise ability. Therefore, this method can be used for the inversion of gravity anomalies.