摘要
针对无人机(UAV)夜间目标检测技术准确率较低的问题,对现有夜间目标检测方法和YOLOv5算法进行简要介绍,提出了基于YOLOv5算法的夜间目标检测技术。通过在原YOLOv5网络中加入改进的Retinex算法对原网络进行动态增强,在模型训练过程中,将Focus层替换成CBS层,改善模型训练效果,并利用改进算法对UAV夜间目标检测性能进行仿真。结果表明,改进的YOLOv5算法在查准率和查全率方面分别比原YOLOv5算法提高了11.22%和5.32%,有效提升了UAV夜间目标检测能力。
Aiming at low accuracy of UAV target detection technology at night,existing night target detection methods and YOLOv5 algorithm were briefly introduced.And a night target detection technology based on YOLOv5 algorithm was proposed.Original YOLOv5 network was dynamically enhanced by adding an improved Retinex algorithm.In the process of model training,Focus layer was replaced by CBS layer to improve model training effect,and the improved algorithm was used to simulate night target detection performance of UAV.Results show that the accuracy and recall ratio of the improved YOLOv5 algorithm are 11.22%and 5.32%higher than that of the original YOLOv5 algorithm respectively,which effectively improves the detection capability of UAV targets at night.
作者
关昊
黎海涛
康衡
杨笑冬
蒋发浩
GUAN Hao;LI Haitao;KANG Heng;YANG Xiaodong;JIANG Fahao(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China)
出处
《无线电通信技术》
2023年第2期345-350,共6页
Radio Communications Technology
基金
北京工业大学星火基金。
作者简介
关昊,北京工业大学本科生。主要研究方向:网络通信与计算。;通信作者:黎海涛,北京工业大学副教授。主要研究方向:智能网络通信与计算。;康衡,北京工业大学本科生。主要研究方向:网络通信与计算。;杨笑冬,北京工业大学本科生。主要研究方向:网络通信与计算。;蒋发浩,北京工业大学本科生。主要研究方向:网络通信与计算。
