U-ISTA Network for Forward-looking Scanning Radar Angular Super-resolution

Scanning radar is of notable importance for ground surveillance, terrain mapping and disaster rescue. However, low angular resolution seriously affects its practical application. The deep neural networks (DNNs) have recently achieved great success in image super-resolution and offered some effective network structures. In this paper, by replacing the two key matrices in the iterative shrinkage thresholding algorithm (ISTA) with learnable weights and incorporating a classical encoder-decoder structure, we propose a U-shaped iterative shrinkage threshold network (U-ISTANet). Specifically, the bottom feature maps in this network are adaptively learned and propagated into subsequent layers, providing an effective way to combine the low-level features and high-level features, and thereby to improve the angular resolution. The simulation results verify the effectiveness of the proposed method.