高地慧眼-ATRNet数据集

数据主编：刘永祥(国防科技大学电子科学学院) 数据集简介：国防科技大学电子科学学院团队旨在推动“高地慧眼-ATRNet”目标样本数据集，为研究者提供良好的数据基准生态。ATRNet-STAR数据集作为团队建立大规模目标特性数据库的第一步，实现了对之前SAR车辆目标基准数据集MSTAR的新突破。团队花费了近两年的时间完成了方案设计、数据采集与处理以及方法基准构建。 ATRNet-STAR收集了来自40种目标类型（涵盖4车辆大类、21种车辆子类、40种车辆型号，包括轿车、SUV、皮卡、客车、货车、罐车等绝大部分民用车辆类型）、丰富场景（包括城区、工厂、林地、裸土和沙地）、各种成像条件（包括不同角度、波段和极化方式）和多种格式（包括浮点型复数原始数据和处理后的8位幅度数据）近20万幅目标图像，附有详细目标尺寸、目标位置、成像角度、分辨率等标注。它是目前最大的公开SAR车辆识别数据集，规模是以往任何车辆数据集的10倍以上。其充足的目标样本可以支撑生成、检测和分类等各方面的研究。 同时，为了便于研究创新和方法比较，团队建立了一个精心设计的分类和检测方法基准ATRBench，包括面向鲁棒识别、少样本识别和迁移学习等7种实验设置和15种代表性方法。实验结果表明，复杂条件下的SAR ATR仍然极具挑战，同时大规模预训练模型表现出了相对优秀的性能，基于该数据集预训练将有助于识别不同地面目标。 该数据集全面的目标样本和实验基准可为SAR ATR提供一个新的研究平台，将为进一步促进SAR ATR领域发展。如有更多需求，欢迎联系邮箱[e-mail] 李玮杰博士。 详细使用说明请参考：高地·慧眼-ATRNet数据集-STAR-1.0-使用说明.pdf 本数据集参考文献与引用格式： [1] 刘永祥，李玮杰，刘丽，周洁，彭渤文，宋娅菲，熊旭颖，杨威，刘天鹏，刘振，黎湘. ATRNet-STAR：面向真实场景遥感目标识别的大规模数据集与基准评测（ATRNet-STAR-1.0）[OL]. 雷达学报, 2025. https://radars.ac.cn/web/data/getData?newsColumnId=f4f66a1b-7924-44ab-9fc2-9ccfbabb27ad Yongxiang Liu, Weijie Li, Li Liu, Jie Zhou, Bowen Peng, Yafei Song, Xuying Xiong, Wei Yang, Tianpeng Liu, Zhen Liu, Xiang Li. ATRNet-STAR: A Large Dataset and Benchmark Towards Remote Sensing Object Recognition in the Wilde (ATRNet-STAR-1.0) [OL]. Journal of Radars, 2025. https://radars.ac.cn/web/data/getData?newsColumnId=f2d2f517-4d2f-4995-8ca5-b2d8d280f862&pageType=en [2] Yongxiang Liu, Weijie Li, Li Liu, Jie Zhou, Bowen Peng, Yafei Song, Xuying Xiong, Wei Yang, Tianpeng Liu, Zhen Liu, Xiang Li. ATRNet-STAR: A Large Dataset and Benchmark Towards Remote Sensing Object Recognition in the Wilde[DB/OL]. (2025-03-13)[2025-03-30]. https://arxiv.org/abs/1706.03825. 发布日期：2025年11月13日

Dataset Editor: Yongxiang Liu (College of Electronic Science and Technology, National University of Defense Technology)

Dataset Introduction: The team from the College of Electronic Science and Technology, National University of Defense Technology developed the "Gaodi Huiyan-ATRNet" target sample dataset to provide a high-quality data benchmark ecosystem for researchers. The ATRNet-STAR dataset serves as the first step in the team's establishment of a large-scale target characteristic database, representing a new breakthrough over the prior SAR vehicle target benchmark dataset MSTAR. The team spent nearly two years completing scheme design, data collection and processing, as well as benchmark method construction.

ATRNet-STAR collects nearly 200,000 target images spanning 40 target categories, which cover 4 major vehicle groups, 21 vehicle subcategories, and 40 vehicle models including most civilian vehicle types such as sedans, SUVs, pickup trucks, buses, freight trucks, tank trucks, etc. The dataset features diverse scenarios (urban areas, factories, woodlands, bare soil, and sand), various imaging conditions (different angles, frequency bands, and polarization modes), and multiple data formats (floating-point complex raw data and processed 8-bit amplitude data). Detailed annotations including target size, target position, imaging angle, and resolution are provided for each sample.

It is currently the largest publicly available SAR vehicle recognition dataset, with a scale more than 10 times that of all previous vehicle datasets. Its abundant target samples can support various research directions including generative modeling, object detection, and classification.

Additionally, to facilitate research innovation and method comparison, the team has developed a well-designed classification and detection benchmark named ATRBench, which includes 7 experimental setups such as robust recognition, few-shot recognition, and transfer learning, as well as 15 representative baseline methods. Experimental results demonstrate that SAR ATR under complex conditions remains highly challenging, while large-scale pre-trained models exhibit relatively strong performance. Pre-training based on this dataset will aid in the recognition of various ground targets.

With its comprehensive target samples and standardized experimental benchmark, this dataset provides a novel research platform for SAR ATR, and will further advance the development of the SAR ATR field. For further inquiries, please contact Dr. Weijie Li at [e-mail]. For detailed usage instructions, please refer to "Gaodi Huiyan-ATRNet Dataset-STAR-1.0-User Guide.pdf".

References and citation formats for this dataset:
[1] Yongxiang Liu, Weijie Li, Li Liu, Jie Zhou, Bowen Peng, Yafei Song, Xuying Xiong, Wei Yang, Tianpeng Liu, Zhen Liu, Xiang Li. ATRNet-STAR: A Large Dataset and Benchmark Towards Remote Sensing Object Recognition in the Wild (ATRNet-STAR-1.0) [OL]. Journal of Radars, 2025. https://radars.ac.cn/web/data/getData?newsColumnId=f2d2f517-4d2f-4995-8ca5-b2d8d280f862&pageType=en
[2] Yongxiang Liu, Weijie Li, Li Liu, Jie Zhou, Bowen Peng, Yafei Song, Xuying Xiong, Wei Yang, Tianpeng Liu, Zhen Liu, Xiang Li. ATRNet-STAR: A Large Dataset and Benchmark Towards Remote Sensing Object Recognition in the Wild[DB/OL]. (2025-03-13)[2025-03-30]. https://arxiv.org/abs/1706.03825.

Release Date: November 13, 2025