FISTA-LPW: Super-Resolution Imaging of Low- Frequency Radar Based on Sparse Regularization with Learnable Partition Weights

This paper introduces a novel approach for bandwidth extrapolation (BWE) imaging of low-frequency radar based on sparse regularization with learnable partition weights (LPWs). The proposed method enhances the quality of super-resolution image reconstruction with limited observation bandwidth through weighting strategies and supervised learning. First, we assign distinct LPWs to scattering centers based on their frequency dependence factors (FDFs), allowing for different regularization strengths that improve BWE accuracy. We then propose FISTA-LPW, an unrolling algorithm designed to optimize these LPWs via supervised learning, utilizing full band spectrums as training labels to minimize extrapolation error in a data-driven manner. Our analysis demonstrates that sparse regularization with LPWs effectively addresses degraded extrapolation accuracy in scenarios with limited observation bandwidth. Experimental results from both simulated and real radar data show that our method outperforms existing BWE techniques in terms of super-resolution image quality, achieving a radar image reconstruction accuracy of 90% while using only 75% of the full bandwidth on real radar data.