Deblurring of radar images aided by point spread function estimator and convolutional regulariser

This work introduces a deep learning-based approach for mitigating motion blur in radar images, caused by either radar platform or target movement. Utilising convolutional neural networks (CNNs), the method learns to map blurred images to their sharp counterparts. Additionally, a separate CNN estimates the point spread function (PSF) of the motion blur, which is then utilised to reconstruct deblurred images. The reconstruction process is further refined by integrating the input image, PSF and ground truth relationship into the training loss term. Trained on a comprehensive dataset of simulated blurred and sharp radar images, the method is evaluated across varying degrees and lengths of blur, surpassing state-of-the-art methods in both qualitative and quantitative assessments. With its superior performance, the proposed approach holds promise for enhancing various radar imaging applications, including target detection, recognition, surveillance, and navigation.