High-Resolution ISAR Image Generation Using Deep Learning with Limited Frequency Bands

In this paper, we propose a deep learning-based model to generate high-resolution inverse synthetic aperture radar (ISAR) images under a limited radar bandwidth. While traditional ISAR systems achieve a higher resolution with a wider bandwidth, this comes at the cost of increased system resource consumption. To address this issue, we propose a method to generate high-resolution images from low-resolution ISAR images using deep learning. The training data used in this study consist of ISAR image pairs generated by calculating the scattered field using physical optics. The images generated by the proposed model can achieve average performances of 0.08, 8.94, and 0.19 in terms of learned perceptual image patch similarity, natural image quality evaluator, and deep image structure and texture similarity, respectively, compared to the original images.