Deep Learning for Radar Automatic Target Recognition: Past, Present, and Emerging Techniques

As modern Artificial Intelligence / Machine Learning (AI/ML) techniques are applied to radar sensor design (e.g., cognitive radar) to collected data analysis (e.g., image formation, detection and classification, and intelligence assessment), several pivotal technical issues remain to be explored. Among these issues, one of them is achieving nearly closed form solutions (such as automatic target recognition, ATR) by deep learning models. Analogous to maximum likelihood estimation (MLE), and maximum a posteriori (MAP) estimation algorithm, we will explore maximum likelihood classification (MLC) to achieve a nearly closed form solution for radar ATR. We will present past (template based) and present ATR techniques (deep learning based). We then discuss a comprehensive list of emerging deep learning techniques to develop an end-to-end ATR system that is built on explainable AI and provides consistent performance in terms of robust and reliable execution. Topics to be covered include active learning, transfer learning, adversarial machine learning, few-shot / low-shot learning, and synthetic radar data. A suggested textbook for this lecture is “Deep Learning for Radar and Communications Automatic Target Recognition”