SBSR-Net: Multitype Parameters Learning-Based Real Aperture Radar Forward-Looking Superresolution Imaging Framework

Regularization methods are widely employed in real aperture radar forward-looking superresolution imaging. However, these methods suffer from the problem of selecting multi-type parameters. To mitigate this issue, based on the real aperture radar forward-looking superresolution imaging application, we propose a multi-type parameters learning-based Split Bregman superresolution network (SBSR-Net) by establishing a projection relationship between the multi-type parameters of the Split Bregman algorithm and the real beam data. First, before the modular network design, we analyze the impact of various types of parameters on imaging performance in conventional algorithms. Second, based on the characteristics of different types of parameters, including penalty parameters, regularization parameters, threshold parameters, and step parameters, we develop differentiated strategies by formulating them as learnable parameters within network layers. Finally, we construct a dataset for forward-looking superresolution imaging and conduct network training to select these multi-type parameters adaptively. The proposed model-data driven architecture demonstrates superior performance in multi-type parameters selection. Simulations and experiments are given to verify the performance of the proposed framework.