Ground-based Radar Tomography for the Moon Based on ADMM-Net

Ground-based radar offers the advantage of all-weather, all-day operation and has become a significant facility for acquiring high-resolution 3-D images of the Moon. However, traditional tomographic imaging is limited by the observation conditions, particularly in the height dimension. While capable of super-resolution imaging, the existing compressive sensing algorithms struggle with optimizing hyperparameters during the iteration. To address this issue, a ground-based radar tomography super-resolution imaging method for the Moon based on a self-supervised compressive sensing network is proposed. This method redefines the tomographic imaging process as an optimization problem utilizing the alternating direction method of multipliers (ADMM). Subsequently, it extends the ADMM iteration process into a multi-layer neural network with trainable hyperparameters, referred to as ADMM-Net. The validity and robustness of the proposed method are verified through simulation experiments of lunar scenes.