微小型无人机SAR地面小目标检测与识别方法
CSTR:
作者:
作者单位:

1.南京航空航天大学电子信息工程学院,南京 211106;2.南京工程学院计算机工程学院,南京 211167

作者简介:

通讯作者:

朱岱寅,男,教授,博士生导师,E-mail:zhudy@nuaa.edu.cn。

中图分类号:

V262

基金项目:


SAR Ground Small Target Detection and Recognition Methods Via Micro-small UAVs
Author:
Affiliation:

1.College of Electronic and Information Engineering,Nanjing University of Aeronautics & Astronautics,Nanjing 211106, China;2.School of Computer Engineering,Nanjing Institute of Technology,Nanjing 211167,China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    南京航空航天大学(Nanjing University of Aeronautics and Astronautics,NUAA)雷达探测与成像团队利用自主研发的无人机载微小型合成孔径雷达(Synthetic aperture radar,SAR)系统针对不同型号的坦克、装甲车和战机等十余类典型军事目标构建了圆周SAR数据集。通过对多次外场试验数据的高精度成像处理,在多俯仰角单基圆周SAR图像数据集的基础上,扩展了不同双基角组合的双基圆周SAR图像数据集。基于该数据集,本文结合团队在SAR图像目标检测和识别方法及应用方面的研究成果,对基于深度学习的SAR目标检测识别技术进行了回顾和综述,对比了不同神经网络模型在南航无人机载圆周SAR数据集上的检测和识别性能。具体地,在目标检测方面,利用SAR图像固有属性获得目标位置信息并结合单阶段轻量级检测算法,提出利用信息分布规律并结合全局注意力机制捕捉小目标位置信息的检测算法,以提高复杂背景下的小目标检测准确率和效率。在目标识别方面,在通过SAR图像先验信息抑制干扰噪声的基础上,提出利用SAR目标多视角信息联合Transformer的目标识别算法,通过设计视角正则化项以约束多视角之间的关联性从而实现不同视角间的特征融合,提高SAR小目标识别的准确率。从无人机载微型SAR系统对地面目标进行实时检测和识别的实际需求出发,本文还探讨了轻量化检测和识别网络在数字信号处理(Digital signal processing, DSP)平台上的部署方案,同时展示了初步试验结果。最后,本文展望了SAR目标智能检测和识别领域面临的挑战和发展趋势。

    Abstract:

    The radar detection and imaging team at Nanjing University of Aeronautics and Astronautics (NUAA) has developed a circular synthetic aperture radar (SAR) dataset for over ten types of typical military targets, including tanks, armored vehicles, and fighter aircraft, using a self-developed UAV MiniSAR system. Through high-precision imaging processing of multiple field trial datasets, the team expands the original monostatic circular SAR image dataset with multiple elevation angles to include bistatic circular SAR image datasets with varying bistatic angle configurations. Building upon this dataset and integrating the team’s research achievements in SAR image target detection, recognition methodologies, and applications, this paper reviews and summarizes deep learning-based SAR target detection and recognition techniques, comparing the performance of different neural network models on NUAA’s MiniSAR dataset. Specifically, for target detection, the paper proposes to leverage the inherent attributes of SAR images to extract target location information and incorporates a lightweight one-stage detection algorithm. Additionally, an algorithm utilizing information distribution patterns combined with a global attention mechanism is introduced to enhance the detection accuracy and efficiency of small targets in complex backgrounds. For target recognition, after suppressing interference noise using SAR image prior information, the study presents a recognition algorithm that integrates multi-view SAR target information with Transformer architecture. Furthermore, a viewpoint regularization term is designed to constrain interview correlations, which enables effective feature fusion across different perspectives to improve the recognition accuracy of small SAR targets. Addressing the practical requirement of real-time ground target detection and recognition by MiniSAR systems, this paper also explores the deployment strategy of lightweight detection and recognition networks on digital signal processing (DSP) platforms, accompanied by preliminary experimental results. Finally, the challenges and future trends in intelligent SAR target detection and recognition are discussed.

    参考文献
    相似文献
    引证文献
引用本文

朱岱寅,吕吉明,周鹏,俞翔,耿哲,王鹏,陈志成,周涛,叶铮,郭二娜,汤翊钧.微小型无人机SAR地面小目标检测与识别方法[J].南京航空航天大学学报,2025,57(5):781-798

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2025-08-26
  • 最后修改日期:2025-09-15
  • 录用日期:
  • 在线发布日期: 2025-10-27
  • 出版日期:
文章二维码
您是第位访问者
网站版权 © 南京航空航天大学学报
技术支持:北京勤云科技发展有限公司