A Novel Plug-and-Play SAR Reconstruction Framework Using Deep Priors

The reconstruction of synthetic aperture radar (SAR) images from phase history data is an ill-posed inverse problem. Existing reconstruction methods use regularization to tackle the ill-posed nature of the imaging task, while assuming a forward model or trying to learn one. In either case, these methods do not decouple the sensing model and the priors used as regularizers. Recently emerging plug-and-play (PnP) priors is a flexible framework that allows forward models of imaging systems to be matched with the state-of-the-art prior models. Inspired by this, in this work, we propose a novel PnP SAR reconstruction framework. This framework decouples the forward model and the prior model, therefore allows us to replace either of them without affecting the other. In this paper, we demonstrate the use of a convolutional neural network (CNN) based prior model for the reconstruction of synthetic SAR scenes and compare the results with FFT-based and non-quadratic regularization based reconstruction methods. Experimental results show that our framework performs on par or better than with these methods in the majority of the scenarios considered.