Application of U-Net for the Recognition of Regional Features in Geophysical Inversion Results

Engineering geophysical prospecting is used to identify anomalous underground bodies, which are typically identified by inversion results. In recent years, machine learning has been applied to many geophysics studies, including geophysical data processing and fault prediction. Machine learning methods can also be used to identify complex anomalous bodies as direct identification can be difficult due to the complexity of local anomalous bodies. We propose a U-Net network to extract anomalous bodies from inverse results. Our model uses the regional similarity of the stratigraphic structure around the anomalous bodies, and then, the neural network extracts the regional background from the geophysical inversion results. Finally, the local anomalous bodies can be subtracted from the predicted regional results and the original inversion results. We designated a large number of datasets for training and designed triangular models for testing. Resistivity data for contaminated soil in a crane storage yard in Shanghai, China, were then processed and the resistivity range was calculated. We processed resistivity and polarizability data from Yantong Mountain, Heibei, China, to determine the degree of mineralization in the area.