Physics-Driven Deep Learning Inversion for Direct Current Resistivity Survey Data

The direct-current (DC) resistivity method is a commonly used geophysical technique for surveying adverse geological conditions. Inversion can reconstruct the model from data, which an important step in survey. However, inverse problem serious ill-posed that makes it easy to obtain incorrect inversion results. Deep learning (DL) provides new avenues solving problems, and has been widely studied. Currently, most DL methods are purely data-driven depend heavily on labels (real models). real models difficult through field surveys. An network may not be effectively trained without labels. In this study, we built unsupervised scheme based physical law of electric propagation. First, forward modeling process was embedded into training, converted predicted data formed misfit observation data. Unsupervised training independent realized using as loss function. Moreover, dynamic smoothing constraint imposed function alleviate problem. Finally, transfer applied adapt with simulated Numerical simulations tests showed proposed accurately locate depict targets.