Fifty Years of SAR Automatic Target Recognition: The Road Forward

Jie Zhou · 2025

Read Fifty Years of SAR Automatic Target Recognition: The Road Forward by Jie Zhou (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-22159.txt.

Introduction

Jie Zhou and eight co-authors survey fifty years of automatic target recognition in radar images. Synthetic aperture radar can image the ground in nearly all weather and light, which makes it a key source of information about objects and scenes. Recognizing targets in those images automatically has been a central and hard problem in remote sensing for five decades. Their 2025 review, accepted by an IEEE journal, is presented as the first full account of the field.

The survey gathers roughly two hundred sixty research contributions. It covers the core challenges of the task, the important data sets, and the strengths and limits of representative methods. The authors argue that modern AI, especially large models and agents, could transform the field, yet its potential for radar images stays largely locked. They close with future directions, calling for better large-scale data sets, fairer benchmarks, and safe open-source ecosystems.

Catalog record

CREATORJie Zhou
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE88 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-22159

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Fifty Years of SAR Automatic Target Recognition: The Road Forward Corresponding authors: Xiang Li ([email protected]), Yongxiang Liu ([email protected]) and Li Liu ([email protected]). This work was partially supported by the National Natural Science Foundation of China under grant 62376283 and 62531026. Emails: Jie Zhou ([email protected]), Weijie Li ([email protected]), Bowen Peng ([email protected]), Yafei Song ([email protected]) and Gangyao Kuang ([email protected]). Jie Zhou Abstract This paper provides the first comprehensive review of fifty years of synthetic aperture radar automatic target recognition (SAR ATR) development, tracing its evolution from inception to the present day. Central to our analysis is the inheritance and refinement of traditional methods—such as statistical modeling, scattering center analysis, and feature engineering—within modern deep learning frameworks. The survey clearly distinguishes long-standing challenges that have been substantially mitigated by deep learning from newly emerging obstacles. We synthesize recent advances in physics-guided deep learning and propose future directions toward more generalizable and physically-consistent SAR ATR. Additionally, we provide a systematically organized compilation of all publicly available SAR datasets, complete with direct links to support reproducibility and benchmarking. This work not only documents the technical evolution of the field but also offers practical resources and forward-looking insights for researchers and practitioners. A systematic summary of existing literature, code, and datasets are open-sourced at: https://github.com/JoyeZLearning/SAR-ATR-From-Beginning-to-Present . , Yongxiang Liu, Li Liu, Weijie Li, Bowen Peng, Yafei Song, Gangyao Kuang, Xiang Li I Introduction I-A Background Synthetic Aperture Radar Automatic Target Recognition (SAR ATR) integrates remote sensing information processing and computer vision (Fig. 1 ), with the goal of detecting and classifying high-value targets in SAR imagery—an all-weather, day-and-night observation modality unique to Earth Observation (EO) [ 1 , 2 , 3 ] . Since the launch of Seasat-A in 1978, the first spaceborne SAR system, SAR ATR has undergone nearly five decades of evolution, advancing from early statistical scattering models to contemporary deep learning networks and foundation models driven by large-scale data and computational resources [ 4 , 5 , 6 ] . This progress reflects sustained, interdisciplinary efforts across electromagnetics [ 7 ] , signal processing [ 8 , 9 ] , pattern recognition [ 10 ] , and artificial intelligence [ 11 , 12 ] to overcome persistent challenges such as speckle noise, pose sensitivity, dynamic scenes, and data scarcity. Over these fifty years, SAR ATR has demonstrated growing strategic importance in applications including disaster monitoring [ 13 , 14 ] , urban development [ 15 ] , maritime security [ 16 ] , and climate analysis [ 17 ] , establishing itself as a key enabler of intelligent perception technologies [ 18 , 19 ] . Therefore, a comprehensive review of its development is crucial to fostering future theoretical breakthroughs and practical applications. Fig. 1: (a) Most frequent keywords in Web of Science (WoS) core collection remote sensing-related papers from 2020 to 2025. The size of each word is proportional to the frequency of that keyword. We can see that synthetic aperture radar has received significant attention in recent years. (b) Total number of publications related to SAR ATR index in WoS core collection for every two-year period from 1999 to 2024. The surge in publications underscores the escalating importance and increasing research interest in this field. Fig. 2: Timeline milestone of SAR automatic target recognition evolution, including two core tasks of classification and detection, from understanding physics, designing features, learning features to understanding and learning features. ( Classification: SAR [ 20 ] , SAR Imaging [ 21 ] , SAR ATR [ 22 ] , MSTAR [ 23 ] , ASC [ 24 ] , Texture [ 25 ] , SVM [ 26 ] , Fisher-MC [ 27 ] , Adaboost [ 28 ] , Sparse Representation [ 29 ] , SIFT-SAR [ 30 ] , A-ConvNets [ 31 ] , CV-CNN [ 32 ] , WWH [ 33 ] , FEC [ 34 ] , HOG-ShipCLSNet [ 35 ] , CA-MCNN [ 36 ] , PIHA [ 37 ] , VSFA [ 38 ] , EMI-Net [ 39 ] , SARATR-X [ 40 ] . Detection: CFAR [ 41 ] , K Distribution [ 42 ] , Two pa.CFAR [ 43 ] , Go Distribution [ 44 ] , WaveletDet [ 45 ] , FuzzyLogic OSSD [ 46 ] , AOSDS [ 47 ] , SARDet [ 48 , 49 ] , Bi-CFAR [ 50 ] , SSDD [ 51 ] , SER FRCNN [ 52 ] , DAPN [ 53 ] , DBBox-v2 [ 54 ] , FBR-Net [ 55 ] , Centernet++ [ 56 ] , SEFEPNet [ 57 ] , SAR-AIRCraft 1.0 [ 58 ] , DiffDet4SAR [ 59 ] , EarthGPT [ 60 ] , ASC-U2Det [ 61 ] .) I-B Comparison with Previous Surveys Over the past few decades, research related to SAR ATR has made significant contributions to the advancement of this field, with key milestones illustrated in Fig. 2 . Nevertheless, the body of work remains largely fragmented, underscoring the need for a systematic survey that synthesizes progress, identifies core challenges and open problems, and outlines promising future directions. Although there are some reviews on SAR ATR, early reviews have evolved from physics-driven feature engineering [ 48 , 62 ] to task-specific deep-learning retrospectives [ 63 , 64 ] . None has systematically traced the complete fifty-year evolution of the technology. Moreover, prior works have not explicitly addressed how classical physical insights have been inherited, refined, and reformulated within modern deep learning architecture. To clarify the unique contributions of this survey, we summarize related works in TABLE I and highlight the following five distinguishing aspects: (i) Comprehensiveness: This survey covers the entire technical trajectory of SAR ATR, from its statistical origins in the 1970s to contemporary physics-data integrated foundation models in the 2020s. (ii) Inheritance: We analyze how traditional concepts, such as scattering center models and CFAR detection, have been absorbed and reinterpreted within deep network frameworks. (iii) Systematicness: We provide a systematic taxonomy of core challenges in SAR ATR, clearly delineating which have been adequately addressed and which remain open, thereby offering a clear agenda for future research. (iv) Openness: We provide a comprehensive compilation of open-source datasets and code repositories, complete with direct links to facilitate reproducibility and support rapid prototyping. (v) Future-oriented: Building on historical context, we distill key emerging research directions to offer a forward-looking roadmap for the field. I-C Scope and Organization Given the enormous work on SAR ATR shown in Fig. 1 (b), exhaustive coverage within a single article is impractical. Therefore, this survey focuses on: (i) Literature source: peer-reviewed papers from high-impact top journals and conferences relate remote sensing, as well as research with pioneering significance. (ii) Temporal span: Nearly five decades from 1978 (the launch of the first spaceborne SAR satellite, Seasat-A) to the present. (iii) Task scope: target detection and classification in single-channel, static SAR images. For topics such as moving target detection, SAR video target recognition, and polarimetric SAR, this paper lists them as independent research directions in the future. The remainder of this paper is structured as follows. Problem definition, core challenges, and datasets of SAR ATR are summarized in Section II . In Section III , we review the history of SAR

Read the complete text