摘要
传统雷达用于无人机检测时通常存在检测信号较弱且易受杂波干扰而产生漏检错检的现象。如何克服传统雷达检测方法的不足,从监测空域快速检测识别无人机是反无人机攻击研究必须解决的基本问题。针对该问题,在概述无人机飞行特征、常用检测方法和YOLOv11算法基本原理的基础上,提出了一种基于YOLOv11算法的无人机目标检测方法。该方法将上下文锚点注意力机制(CAA)、路径聚合网络(PA-Net)和一个新的检测头引入到YOLOv11算法,实现了对不同重要度特征的选择性加强或抑制,并能融合高级语义特征和多尺度的空间信息来丰富特征的表示。随后,在DUT Anti-UAV公共数据集上对该算法进行了实例验证,结果表明,改进算法相较于原始的YOLOv11算法,在评价指标mAP50和mAP50:90上分别提升了5.1%和7.8%,能有效克服原有算法在小目标检测上的不足,和一些常用算法相比也有一定的优势,其精确度和召回率分别为97.4%和92.5%,能较好地实现对无人机目标的检测。
When traditional radar is used for drone detection,it often suffers from weak detection signals and susceptibility to clutter interference,leading to missed and false detections.Overcoming the limitations of traditional radar detection methods to rapidly detect and identify drones in monitored airspace represents a fundamental challenge that must be addressed in counter-drone attack research.To tackle this issue,a UAV target detection method based on the YOLOv11 algorithm is proposed,building upon an overview of UAV flight characteristics,commonly used detection methods,and the fundamental principles of the YOLOv11 algorithm.The proposed method integrates Context Anchor Attention(CAA)mechanism,Path Aggregation Network(PA-Net),and a novel detection head into the YOLOv11 framework.This integration enables selective enhancement or suppression of features with varying importance levels while effectively fusing high-level semantic features with multi-scale spatial information to enrich feature representation.Subsequently,the algorithm is empirically validated using the DUT Anti-UAV public dataset.The results indicate that,compared to the original YOLOv11 algorithm,the improved version achieved increases of 5.1%and 7.8%in the mAP50 and mAP50:90 evaluation metrics,respectively.This enhancement effectively overcoming the shortcomings of the original algorithm in small target detection.Furthermore,when compared to several commonly used algorithms,the improved method demonstrates certain advantages,achieving a precision of 97.4%and a recall rate of 92.5%.It can achieve better performance in UAV target detection.
作者
肖建国
熊宇虹
张方凯
何伟
舒明磊
Xiao Jianguo;Xiong Yuhong;Zhang Fangkai;He Wei;Shu Minglei(School of Computer Science and Information Engineering,Shanghai Institute of Technology,Shanghai 201418,China)
出处
《航空兵器》
北大核心
2025年第4期80-87,共8页
Aero Weaponry
