基于改进YOLOv5s的无人机航拍小目标检测算法研究
Research on Small Target Detection Algorithm for UAV Aerial Photography Based on Improved YOLOv5s
尹泉贺 1原素慧 1朱梦琳 1兰洁1
作者信息
- 1. 华北水利水电大学 信息工程学院,河南 郑州 450046
- 折叠
摘要
目前,无人机航拍目标检测技术在军事和民用领域得到广泛的应用,但复杂场景中小目标密集,易出现误检和漏检的情况.为此,文章提出一种基于改进YOLOv5s的无人机航拍小目标检测算法,用分组卷积取代两个普通卷积,用解耦检测头取代耦合检测头,去除了原始算法中的P5检测头,在PANet结构中增加一层新的P2检测头.仿真结果表明,改进算法具有较好的检测效果,mAP50较原始算法提高了9.3%,同时能够满足无人机实时性检测需求.
Abstract
At present,UAV aerial photography target detection technology has been widely applied in military and civilian fields,but in complex scenes,small targets are dense and prone to false positives and missed detections.For this purpose,this paper proposes a small target detection algorithm for drone aerial photography based on improved YOLOv5s,which replaces two ordinary convolutions with grouped convolutions and decoupled detection heads instead of coupled detection heads.The P5 detection head in the original algorithm is removed,and a new P2 detection head is added to the PANet structure.The simulation results show that the improved algorithm has good detection performance,with mAP50 increased by 9.3%compared to the original algorithm,and can meet the real-time detection requirements of UAV.
关键词
无人机航拍/小目标检测/空间池化金字塔/解耦检测头Key words
UAV aerial photography/small target detection/spatial pooling pyramid/decoupled detection head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