High-Quality Self-contained Electromagnetic Imaging Scheme Based on Projected Nonlinear Landweber and Machine Learning

This work aims to build an inverse electromagnetic imaging scheme exploiting the best of the following two methodologies: iterative nonlinear regularization and recently developed machine learning models. The proposed method is designed to achieve three targets: 1) the elimination of user intervention to empirically select inverse parameters; 2) high-quality image recovery; and 3) stable scheme with accelerated reconstruction time. To achieve these targets, an iterative imaging algorithm is developed based on projected nonlinear Landweber (PNLW) supported by three neural networks that: 1) predict first-norm solutions, which is an inverse parameter required by the projection operator; 2) predict iteration step lengths that ensure PNLW convergence; and 3) filter recovered images via PNLW to enhance image quality. The first-norm projection operator satisfies the sparse constraint. The proposed scheme is efficient for applications where the number of measurements is limited by adopting sparse optimization. The results of the proposed scheme and conventional approach are compared in numerical tests to assess the efficiency and performance of the proposed scheme.