High-Resolution ISAR Imaging and Autofocusing via 2D-ADMM-Net

A deep-learning architecture, dubbed as the 2D-ADMM-Net (2D-ADN), is proposed in this article. It provides effective high-resolution 2D inverse synthetic aperture radar (ISAR) imaging under scenarios of low SNRs and incomplete data, by combining model-based sparse reconstruction data-driven deep learning. Firstly, mapping from ISAR images to their corresponding echoes wavenumber domain derived. Then, a alternating direction method multipliers (ADMM) unrolled generalized network, where all adjustable parameters layers, nonlinear transform multiplier update layers are learned an end-to-end training through back-propagation. Since optimal each layer separately, 2D-ADN exhibits more representation flexibility preferable performance than model-driven methods. Simultaneously, it able better facilitate with limited samples methods owing its simple structure small number parameters. Additionally, benefiting good 2D-ADN, random phase error estimation proposed, which well-focused can be acquired. demonstrated experiments that although trained only few simulated images, shows adaptability measured data favorable results clear background obtained short time.