Review of Synthetic Aperture Radar Automatic Target Recognition: A Dual Perspective on Classical and Deep Learning Techniques

With advances in both classical and modern methods,synthetic aperture radar (SAR) automatic target recognition (ATR) has made enormous progress. This article reviews a wide range of SAR ATR techniques, from model- and template-based approaches to statistical techniques, such as constant false alarm rate, as well as the increasing impact of machine learning and deep learning. What makes SAR imagery particularly unique are problems, such as speckle noise, target variability, and clutter, for which there are specialized solutions described in this article. Feature-based approaches in traditional SAR ATR followed from conventional feature-driven strategies, while modern data-driven end-to-end recognition methods, consisting primarily of convolutional neural network based and hybrid networks, have been applied. The performance has further been enhanced with techniques, such as transfer learning, unsupervised learning, and adversarial learning, to overcome data scarcity and variability. Alongside these techniques, this review also looks at application-specific methods suited to operational needs, such as real-time detection, robust classification, an identification of small objects, and new data handling techniques, such as data augmentation and multimodal fusion. Based on architectures, learning paradigms, and operational contexts, a detailed taxonomy bridges classical and contemporary SAR ATR methods. This article consolidates advancements and outlines challenges to present a unified framework for researchers and practitioners to understand future directions of SAR ATR development in military and civilian applications.