Stable downward continuation of the gravity potential field implemented using deep learning

Downward continuation (DC) of the gravity potential field is an important approach used to understand and interpret density structure boundary anomalous bodies. It widely delineate highlight local shallow sources. However, it well known that direct DC transformation in frequency domain unstable easily affected by high-frequency noise. Recent deep learning applications have led development image recognition resolution enhancement using convolutional neural network technique. A similar architecture also suitable for training a model problem. In this study, solve problems existing methods, we constructed dedicated called DC-Net We fully trained on 38,400 pairs anomaly data at different altitudes network. conducted several experiments implemented real-world example. The results demonstrate following. First, validation subset test prediction indicate was sufficiently trained. Moreover, performed better than traditional strategy refining upscaling low-resolution images. Second, tests datasets with changing levels noise demonstrated noise-resistant robust. Finally, proposed example, which demonstrates solving problem delineating detailed feature near source. For real processing, should be reduced advance. Additionally, recommend quantification before training.