基于YOLOv8的无人机编队领航者检测算法
UAV Formation Leader Detection Algorithm Based on YOLOv8
黄祎闻 1甄子洋 1何佳璐1
作者信息
- 1. 南京航空航天大学自动化学院,江苏南京 211106
- 折叠
摘要
基于视觉的无人机编队方法具有不受通信拒止影响的优点,与传统编队算法相比有更强的鲁棒性,逐渐成为了领域内的研究热点.在Leader-Follower无人机视觉编队模式中,跟随者通过对领航者执行实时目标检测,并解算出领航跟随者之间的相对位置关系来完成编队控制任务.基于YOLOv8n目标检测模型提出了 一种改进的实时目标检测算法:在Neck模块中加入可变形卷积模块;加入多头注意力机制增强特征提取;在训练过程中进行数据增强.为验证所提算法的性能优势,进行了 2次对比测试,实验结果表明,改进算法比原始算法的特征提取效果更强,检测精度更高.最后,将改进的领航者检测算法应用于无人机编队任务中,证明了所提算法的实际应用价值.
Abstract
The vision-based unmanned aerial vehicle(UAV)formation method has the advantage of being unaffected by communication disruptions and exhibits greater robustness compared to traditional for-mation algorithms,gradually becoming a research hotspot in the field.In the Leader-Follower UAV visual formation mode,followers achieve formation control by performing real-time target detection on the lead-er and calculating the relative positional relationship between the leader and the followers.This paper pro-poses an improved real-time object detection algorithm based on the YOLOv8n object detection model:convolution modules were added in the Neck module,a multi-head attention mechanism was added to en-hance feature extraction,apply data augmentation was applied in the training process.To validate the per-formance advantages of the algorithm proposed,two comparative tests were conducted.The experimental results indicate that the improved algorithm exhibits stronger feature extraction and higher detection accu-racy compared to the original algorithm.Finally,the improved object detection algorithm is applied to drone formation tasks,demonstrating the practical utility of the algorithm in this context.
关键词
无人机编队/YOLOv8/可变形卷积/多头自注意力机制Key words
UAV formation/YOLOv8/deformable convolution/multi-head attention mechanism引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据