Trained Electromagnetic Imaging Approach based on Born Iterative Method

A trained regularization approach is proposed for two-dimensional (2D) electromagnetic imaging based on the Born iterative method (BIM). The algorithm reconstructs the material's real permittivity profile after applying a trained version of BIM (TBIM). TBIM starts by reconstructing the permittivity profile using the Born Approximation. Then a loop is executed that mutually retrieves the total fields through matrix inversion, and the permittivity, though optimizing an ill-posed linear problem. The ill-posed problem was solved using a Trained-Iterative Shrinkage Thresholding Algorithm (TISTA). TISTA consists of linear Landweber iteration and trained U-net regularization networks. The numerical results shown in this work demonstrate high-quality image restoration.