Computational Radar Imaging
Friday, October 18, 2019 - 9:05am - 9:50am
We present synthetic aperture radar (SAR) as a computational imaging modality, emphasizing aspects of radar that differentiate it from other imaging problems. In this context, we present samples of work resulting from two related lines of inquiry in our group: (1) sparsity-driven radar imaging, and (2) machine learning for radar imaging. Specific pieces of the first line of work include (i) analysis and synthesis-based sparse signal representation formulations for SAR image formation together with the associated algorithms and imaging results; (ii) sparsity-based methods for wide-angle SAR imaging and anisotropy characterization; (iii) sparsity-based methods for joint imaging and autofocusing from data with phase errors; and (iv) techniques for exploiting sparsity for SAR imaging of scenes containing moving objects. Next, we turn to more recent work that brings in machine learning into the process of SAR image formation. As an initial attempt, we describe dictionary learning methods aimed to tune sparsity-driven priors to a particular context. Then, we present a framework in which we incorporate convolutional neural network (CNN) based prior models into SAR image formation. In particular, we propose a plug-and-play (PnP) priors based approach that allows joint use of physics-based forward models and state-of-the-art prior models. We demonstrate preliminary results demonstrating the potential of this machine learning based approach for the reconstruction of synthetic and real SAR scenes.