Advances in Multi-Resolution Approaches for Computational Inverse Scattering - On the Integration of Sparse Retrieval within the Multi-Resolution Inversion

In this paper, an innovate approach which combines a customized sparseness-regularized solver with a multi-scaling procedure for the reconstruction of sparse two-dimensional (2D) dielectric profiles is presented. A customized fast Relevant Vector Machine (RVM), constrained to estimate the sparse unknown coefficients only within a restricted research space defined according to the information progressively acquired during the multi-scaling procedure, is used to solve the inverse problem formulated as a Bayesian Compressive Sensing (BCS) one. Selected numerical results are presented in order to numerically validate the proposed method also in a comparative assessment with the bare approach.