Rotational Motion Compensation and Sparse ISAR Imaging of Maneuvering Targets via Deep Unrolling Network

Accurate rotational motion compensation is critical for achieving well-focused inverse synthetic aperture radar (ISAR) imaging of maneuvering targets. However, low signal-to-noise ratio (SNR) and incomplete echoes often lead to significant performance degradation in conventional methods. Furthermore, these methods rely heavily on manual parameter tuning, which limits their adaptability to varying SNR and data missing rate in practical applications. In this article, a novel deep unrolling network for ISAR imaging of maneuvering targets is proposed. Firstly, an iterative method, termed RMC-PDHG, is proposed for rotational motion compensation and well-focused ISAR imaging based on Primal-Dual Hybrid Gradient (PDHG), enabling accurate imaging of maneuvering targets under low SNR and incomplete echo conditions. On this basis, a rotational motion compensation and imaging network, i.e., RMC-PDHG-Net, is developed by unrolling the RMC-PDHG. This network incorporates a hypernetwork to dynamically generate optimal internal parameters such as the regularization coefficient and step size based on intermediate image features, thereby gaining robustness to varying SNR and data missing rate. Additionally, a two-stage training strategy combining unsupervised and supervised learning is proposed to improve rotation parameter estimation accuracy and image reconstruction quality. Experimental results on simulated and measured data have demonstrated the effectiveness and robustness of the proposed network.