Single-Transmitter Electromagnetic Inverse Scattering Based on Field Continuation and Local Inversion

Inverse Scattering Problems (ISPs) deduce an object's physical properties non-intrusively, which are important in diverse fields from biomedical imaging to geophysical exploration. Most existing inverse scattering methods rely on multiple transmitters and dense receivers. However, when only a single transmitter is available, the incident wave leads to fewer solution constraints and ill-posed inversion. This letter introduces an F2N-N2P Net, a novel deep learning framework that decomposes the inversion into two physically-consistent subtasks: field continuation and local inversion. First, a far-field to near-field network (F2N-Net) recovers near-field data from far-field measurements. By converting the direct inverse mapping to homomodal field continuation, it provides more information constraints for the ill-posedness problem. Second, a near-field to permittivity network (N2P-Net) robustly reconstructs target permittivity through predicting contrast current from the dense near-field data. Experiments demonstrate that our framework effectively reconstructs the target's permittivity.