摘要
针对目前在无人机图像目标检测算法中存在漏检与误检、不能兼顾检测速度,并且不能很好地应用于移动设备端等问题,提出了以YOLOv5n算法为基础进行改进的无人机图像目标检测算法。在原有的网络结构中添加小目标检测层M,增强对小目标的检测能力;在主干特征提取网络中引入BoT模块,减少网络参数量计算并提高检测精度;在特征融合网络中添加CBAM注意力机制,有效抑制背景信息干扰;将网络的头部替换成解耦头部,增强网络的收敛效果。将改进的算法在处理后的VirDrone数据集上进行测试,实验结果表明,在YOLOv5n算法上整体平均精度均值提升了10.25%,检测精度提高了9.81%,改进后的算法在保证实时性的同时有效提高了检测精度。
In view of the current missed and false detection in the UAV image target detection algorithm,the detection speed cannot be taken into account,and it cannot be well applied to mobile devices.An improved UAV image target detection algorithm based on YOLOv5n algorithm is proposed.Firstly,the small target detection layer M is added to the original network structure to enhance the detection ability of small targets.Secondly,the BoT module is introduced in the backbone feature extraction network to reduce the calculation of network parameters and improve the detection accuracy.The CBAM attention mechanism is added to the feature fusion network to effectively suppress background information interference.Finally,the head of the network is replaced with the decoupling head to enhance the convergence effect of the network.The improved algorithm is tested on the processed VirDrone dataset.The experimental results show that the overall average accuracy mean on the YOLOv5n algorithm is improved by 10.25%and detection accuracy improved by 9.81%.The improved algorithm effectively improves the detection accuracy while ensuring real-time performance.
作者
罗旭鸿
刘永春
楚国铭
蒲红平
LUO Xuhong;LIU Yongchun;CHU Guoming;PU Hongping(School of Automation and Information Engineering,Sichuan University of Science&Engineering,Zigong 643002,China;Artificial Intelligence Key Laboratory of Sichuan Province,Yibin 644005,China;School of Unmanned Aerial Vehicles Industry,Chengdu Aeronautic Polytechnic,Chengdu 610100,China)
出处
《无线电工程》
北大核心
2023年第7期1528-1535,共8页
Radio Engineering
基金
人工智能四川省重点实验室科研项目(2020RZY01)
厅市共建智能终端四川省重点实验室项目(SCITLAB-20011)
四川轻化工大学研究生创新基金资助项目(Y2022161)。
