Multifrequency Electromagnetic Inversion Based on Early Fusion Deep Learning Framework

This paper utilizes the early fusion deep learning framework (EFF) to address real-time multifrequency electromagnetic (EM) inverse scattering problems (ISPs) with heightened precision. The EFF incorporates both the scattered field encoder (SFE) and the backpropagation encoder (BPE). his fusion involves amalgamating the extracted features of the scattered field and the BP method across various frequencies, thereby capturing multiple scattering phenomena more comprehensively during the solving process. Additionally, it has been observed that the mutual regularization of multiple scattering information and BP distributions effectively enhances the encoding process, leading to accelerated convergence and more stable inversion results. Moreover, the multifrequency data provide more adequate information in the solving process, and effectively improve the performance of EFF. The superior veracity, stability and generalization ability of the multifrequency EFF have been demonstrated by the synthetic results.