3D Joint Inversion of Multi-physics Data Using Deep Learning Techniques

Joint inversion of multi-physics data can reduce inversion uncertainties and improve resolution. In this abstract, an efficient approach is proposed for 3Djoint inversion using a deep learning network to generate initial models for individual inversions. A two-round iteration strategy is adopted to improve the reasonability of the recovered models and the overall efficiency of the inversion scheme. Experiments demonstrate that our method can achieve higher accuracy and efficiency, compared to individual inversions and cross-gradient-based joint inversion.