基于改进YOLOv7的反无人机目标识别技术的研究
Research on anti-UAV target recognition technology based on improved YOLOv7
梅礼坤 1陈智利 1李栋琦1
作者信息
- 1. 西安工业大学光电工程学院,西安 710021
- 折叠
摘要
无人机的广泛应用和军事化给国家和社会安全带来了及其严重的危害,针对传统反无人机系统设备在民事部署方面存在立法瓶颈,而且缺乏多无人机目标同时检测和识别机制的问题,提出了一种基于YOLOv7的无人机目标识别技术.利用YOLOv7网络对高空多场景下无人机目标进行识别:在YOLOv7中引入Feature reuse based on concatenation模块,解决backbone部分特征重用有限和深度网络信息丢失的问题;使用ELAN of attention mechanism模块提高特征融合时去除噪声和抑制不相关信息的能力;利用HEAD of expansive convolution and residual theory模块降低小目标无人机漏检的问题.在同一参数和数据集下进行训练,结果表明:与YOLOv7原始模型相比,改进模型的mAP@0.5提高了 2.8％,解决了原始网络针对细小目标漏检的问题,弥补了反无人机在民事应用中的不足.
Abstract
The widespread application and militarization of UAVs have brought serious harm to the national and so-cial security.Aiming at the problems of legislative bottleneck in the civil deployment of traditional anti-UAVs system equipment and the lack of simultaneous detection and recognition mechanism of multiple UAVs,a UAVs target recogni-tion technology based on YOLOv7 was proposed.Using YOLOv7 network to identify UAV targets in high-altitude multi-scene environment:a Feature reuse based on concatenation module is introduced in YOLOv7 to solve the problems of limited feature reuse in backbone and information loss in deep network.ELAN of attention mechanism module is used to improve the ability of removing noise and suppressing irrelevant information in feature fusion.The HEAD of expan-sive convolution and residual theory is used to reduce the problem of missing detection of small target UAVs.The re-sults show that compared with the original YOLOv7 model,the average accuracy of the improved model is increased by 2.8％,which solves the problem of missing detection of small targets in the original network and makes up for the shortcomings of anti-UAV in civil application.
关键词
特征复用/自适应机制/细小目标Key words
feature reuse/adaptive mechanism/small target引用本文复制引用
基金项目
国家级科研项目(G20210101)
陕西省科技厅(2023-YBGY-369)
出版年
2024
NETL
NSTL
万方数据