Iterative Multiresolution Bayesian CS for Microwave Imaging

Abstract

A new compressive sensing (CS) imaging method is proposed to exploit, during the inversion process and unlike state-of-the-art CS-based approaches, additional information besides that on the target sparsity. More specifically, such an innovative multiresolution (MR) Bayesian CS scheme profitably combines: 1) the a priori knowledge on the class of targets under investigation and 2) the progressively acquired information on the scatterer location and size to improve the accuracy, the robustness, and the efficiency of both standard (i.e., uniform-resolution) CS techniques and MR/synthetic-zoom approaches. Toward this end, a new MR-based information-driven relevance vector machine (RVM) is derived and implemented. Selected results from an extensive numerical and experimental validation are shown to give the interested readers some indications on the effectiveness and the reliability of the proposed method also in comparison with state-of-the-art deterministic and Bayesian inversion techniques.