基于改进YOLOv5的无人机目标识别方法
A Method for UAV Object Recognition Based on Improved YOLOv5
魏麟 1何峻毅 2谭任翔 2彭俊榕2
作者信息
- 1. 中国民用航空飞行学院 飞行技术学院,四川 德阳 618307
- 2. 中国民用航空飞行学院 航空电子电气学院,四川 德阳 618307
- 折叠
摘要
据统计资料显示,截至2017年全球已经售出无人机达300万架.无人机具有体积小、成本低、数量大等特点,这也随之引起了一系列的安全问题,对公共安全造成了严重威胁.传统的对无人机的识别方法主要有雷达探测、声波探测等.在分析了传统识别方法的劣势后,提出了基于改进YOLOv5 的无人机识别方法,在原始YOLOv5 模型的基础上添加CBAM注意力机制,以增强目标特征提取的能力,增强网络模型性能.并可引入DeepSORT跟踪算法,为无人机的跟踪提供检测响应.经过测试集测试,改进后的模型较原始模型的精确度提升了5.24%,基本满足识别要求.
Abstract
According to statistical data,as of 2017,over 3 million UAVs have been sold worldwide.UAVs have characteristics such as small size,low cost,and large quantity,which lead to a series of security issues and pose a serious threat to public safety.The traditional methods for UAV recognition mainly include radar detection and acoustic detection.After analyzing the drawbacks of these traditional recognition methods,this paper proposes an UAV recognition method based on improved YOLOv5.It incorporates the CBAM(Convolutional Block Attention Module)Attention Mechanism based on the original YOLOv5 model,so as to enhance the capability of target feature extraction and improve the performance of the network model.Furthermore,the DeepSORT tracking algorithm can be introduced to provide detection response for UAV's tracking.Through testing on the dataset,the accuracy of the improved model is improved by 5.24%compared to the original model,which meets the recognition requirements basically.
关键词
无人机/目标识别/YOLOv5/CBAM/DeepSORTKey words
UAV/object recognition/YOLOv5/CBAM/DeepSORT引用本文复制引用
基金项目
四川省通用航空器维修工程技术研究中心资助课题(GAMRC2021YB11)
民航飞行技术与飞行安全重点实验室自主研究项目(FZ2021ZZ05)
出版年
2024
国家科技期刊平台
NSTL
万方数据