An Effective Method Incorporating Binary Prior Information for Programmable Metasurface-Based Microwave Computational Imaging

Abstract

Microwave computational imaging (MCI) combined with programmable metasurface (PMS) has seen significant advancements in recent years. This new microwave imaging technology performs multiplexed measurements by manipulating the radiation pattern of PMS and acquires the spatial resolution. Compared with the traditional real aperture microwave imaging and synthetic aperture microwave imaging, PMS-based MCI (PMS-MCI) not only reduces the cost of the imaging system, but also significantly improves imaging efficiency. As a typical inverse scattering problem, PMS-MCI is nonlinear. To address this nonlinearity, the Born approximation or physical optical (PO) approximation is often used. Additionally, the limited number of independent PMS radiation patterns makes PMS-MCI an ill-posed problem. The ill-posedness of PMS-MCI is mostly overcome through a regularisation scheme which leverages sparse prior information. However, the imaging performance of these existing sparsity-regularised methods can degrade significantly if the sparsity of the probed scene decreases. In some scenarios, one only seeks to reconstruct the shape of a metallic object, which can be parameterised with a binary local shape function (LSF). This binary prior information of LSF can also be exploited to tackle the ill-posed problem. Therefore, a method incorporating such a priori binary information will be introduced into PMS-MCI for recovering the shape of metallic objects in this work. Specifically, a prior model is first constructed to enforce the binary characteristics of the unknowns. Then, Bayesian inference is performed using the variational expectation maximisation (EM) algorithm, integrated with the damped generalised approximation message passing (GAPM) algorithm. Numerical examples are presented to demonstrate the accuracy, efficiency and robustness of the proposed PMS-MCI method.