Leveraging Radar Back-Scattered Data for Classification of Imaging Targets

Employing deep learning methodologies for computer vision tasks, particularly in the domain of radar image analysis, necessitates access to a large and diverse dataset. In the context of radar imagery, the creation of such a dataset often entails the intricate task of reconstructing images from raw radar back-scattered data. This reconstruction process involves handling substantial data volumes, which can be computationally intensve and time-consuming. In this research, a deep learning framework is proposed for target classification utilizing solely the radar back-scattered data, completely bypassing the need for image reconstruction procedure, thereby significantly reducing the classification time. To make the dataset generation easier, a computational imaging (CI) numerical model is employed. Subsequently, the deep learning model is trained using this dataset, and following the training phase, it is tested with radar back-scattered data that is not included in the network training. The outcomes of this evaluation confirm the benefit of training a deep learning model to perform image identification tasks based on radar back-scattered signatures.