MF-JMoDL-Net: A Sparse SAR Imaging Network for Undersampling Pattern Design Toward Suppressed Azimuth Ambiguity

Breaking the constraint of pulse repetition frequency (PRF) is one of the important development trends of synthetic aperture radar (SAR). Within the conventional azimuth sampling patterns, severe ambiguity arises when confronted at a low PRF. Conversely, elevated PRF introduces considerable data redundancy, thereby culminating in the wasting of resources. To address these issues, this article proposes a novel joint optimization network for sparse SAR imaging and azimuth undersampling pattern grounded in the model-based deep learning (MoDL) priors architecture, combined with matched filter (MF) approximate measurement operators, named MF-based sampling pattern joint optimization MoDL sparse SAR imaging Network (MF-JMoDL-Net). The MF-JMoDL-Net incorporates nonuniform sampling operators, enabling the sampling positions to be learnable and achieving the groundbreaking joint optimization of the sampling pattern and ambiguity suppression. When the PRF is below the Nyquist sampling rate, the proposed network can acquire SAR images with minimal ambiguity and optimal imaging quality. Furthermore, the final learned undersampling pattern can be visualized and combined with the SAR echo signal semantics for mutual feedback. Extensive experiments on simulated and real scenes datasets are conducted to demonstrate the effectiveness and superiority of the proposed framework in imaging results.