Automatic Target Recognition (ATR) in Ground Surveillance Radar (GSR) systems is of paramount importance due to its critical applications across defense, security, and civilian domains. This paper provides a comprehensive review of the evolution of ATR systems, tracing their progression from foundational, manually engineered methods to modern advanced techniques, with a particular focus on machine learning (ML) and deep learning (DL) approaches. It analyzes a wide spectrum of features utilized for target classification, ranging from conventional descriptors—such as mean Radar Cross-Section (RCS) and micro-Doppler signatures—to complex representations automatically extracted by deep learning architectures, such as autoencoders. Furthermore, various target classification methodologies and major domain-specific challenges, including the scarcity of large-scale, high-quality datasets, are evaluated. Finally, a comparative analysis of different ATR techniques is presented in terms of classification performance, computational complexity, and implementation constraints, concluding with a discussion on emerging trends and future research directions in this field.