基于改进YOLOv5的飞机目标检测算法  

Aircraft target detection algorithm based on improved YOLOv5

作　　者：张贝贝 刘建辉[1] 王鑫[1] 魏祥坡 麻顺顺 ZHANG Beibei;LIU Jianhui;WANG Xin;WEI Xiangpo;MA Shunshun(Institute of Data and Target Engineering,Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学数据与目标工程学院,河南郑州450001

出　　处：《海洋测绘》2025年第1期66-71,共6页Hydrographic Surveying and Charting

基　　金：国家自然科学基金项目(41801388);河南省自然科学基金项目(222300420386)。

摘　　要：由于飞机目标的大小和形状差异、遮挡、密集分布以及复杂背景等因素,现有模型在检测过程中存在较多的错检和漏检。为此,提出一种基于改进YOLOv5的飞机目标检测模型,称为AT YOLOv5。首先在主干网络中融合坐标注意力模块,增强模型的特征提取能力,然后针对FPN在特征融合时多尺度表征能力降低的问题,提出了注意力特征融合网络,该结构基于注意力权重可以实现不同尺度特征的自适应融合。最后,改进小目标检测层,并在所有检测层中加入Swin Transformer模块,以增强网络模型获取全局信息和关联目标信息的能力。实验部分采用DOTA和RSOD数据集来验证模型的有效性及泛化能力。实验结果表明,提出的检测算法在DOTA数据集下AP 50相对于YOLOv5s网络提高了3.9%,AP 50:95提高了1.0%,对高分辨率遥感影像的FPS可达到64,在RSOD数据集下的AP 50也可以达到96.7%。本文算法可以有效实现飞机目标检测任务,具有较好的检测精度、实时性和鲁棒性。Due to the size and shape differences of aircraft targets,occlusion,dense distribution and complex background,the existing models have high error and missed detection in the detection process.To this end,this paper proposes an aircraft target detection model based on an improved YOLOv5,called AT YOLOv5.Firstly,the coordinate attention module is integrated in the backbone network to enhance the feature extraction ability of the model.Then,aiming at the problem that the multi-scale representation ability of FPN is reduced during feature fusion,an attention feature fusion network is proposed,which can realize the adaptive fusion of features of different scales based on attention weight.Finally,the small target detection layer is improved,and the Swin Transformer module is added to all the detection layers to enhance the ability of the network model to obtain global information and associate target information.In the experimental part,DOTA and RSOD datasets are used to verify the effectiveness and generalization ability of the model.The experimental results show that the AP 50 of the detection algorithm proposed in this paper is 3.9%higher than that of the YOLOv5s network under the DOTA dataset,and the AP 50:95 is 1.0%higher than that of the YOLOv5s network.The FPS of high-resolution remote sensing images can reach 64,and the AP 50 of the detection algorithm under the RSOD dataset can also reach 96.7%.The proposed algorithm can effectively realize the aircraft target detection task,and has good detection accuracy,real-time performance and robustness.

关 键 词：飞机目标检测 深度学习 YOLOv5网络 注意力机制优化 特征融合 

分 类 号：P23[天文地球—摄影测量与遥感]