Deep Learning-Based SAR Image Generation Under Non-Uniform Vehicle Speed

In this paper, we propose a convolutional neural network (CNN)-based method to compensate for distortions that can occur when acquiring radar data from a synthetic aperture radar (SAR) system. The SAR system is widely used in various fields, including terrain observation, disaster monitoring, surveillance and reconnaissance. However, if the platform carrying the radar sensor moves at a non-uniform speed, distortions appear in the SAR image along the direction of the speed change. Therefore, we propose a CNN-based compensation method to address these distortions, and we evaluate its performance. Experiments are conducted in two distinct environments by mounting a radar sensor on a moving platform and varying the speed of the platform. The range-Doppler algorithm is then applied to the radar data to generate SAR images. Then, we use the CNN-based model to compensate for distortions in SAR images acquired while the platform was moving at varying speeds. The performance of the proposed method is evaluated by comparing SAR images captured at a constant platform speed with SAR images before and after compensation. The overall results obtained in three distinct environments revealed consistent gains. Using the proposed method, we achieved a 11.28%p increase in peak signal-to-noise ratio (PSNR) and a 1.92%p improvement in structural similarity index measure (SSIM) in Environment 1. In Environment 2, the method yielded a 20.46%p increase in PSNR and a 3.25%p increase in SSIM. For Environment 3, the results were a 5.88%p increase in PSNR and a 1.55%p increase in SSIM.