Deep Unfolding of the Half-Quadratic Splitting Algorithm for ISAR Image Super-resolution
Zhixiong Yang, Jingyuan Xia, Tianrui Liu, Shuaifeng Zhi, Zhen Liu · 2021 CIE International Conference on Radar (Radar) · 2021
This paper targets on an effective and interpretable model-based deep learning method for Inverse Synthetic Aperture Radar (ISAR) image super-resolution. Due to the intrinsic hardware limitations of the radar system, ISAR images are typically blurred and of low-resolution which lead to difficulties in down-streaming tasks including target detection, recognition and classification. The traditional model-based methods would be less effective when competing with the data-driven meth-ods. However, these deep learning model seldom has a good interpretative formulation to be studied mathematically. In this paper, we propose an unfolded half-quadratic splitting network (UHQSNet) that establishes a hierarchical network by unfolding the iterative optimization process into the fine-designed network architecture. The experimental results on a three-target ISAR datasets show that the proposed UHQSNet gains significantly better performance than both of the model-based method and deep learning method.