| 摘要: |
| 为解决由于空中多目标及其多尺度特征导致的目标检测召回率低、精确度低等问题，提出了一种基于YOLOv7改进的无人机多目标光学检测方法。针对无人机蜂群目标，使用K-means算法对自制的多尺度无人机数据集（Multi Scale Drone Dataset，MSDD）优化聚类，对原有YOLOv7锚框进行增改；在特征融合网络部分加深网络层数，使更深层的特征与浅表特征进一步融合，增强小尺度目标的特征表达能力；在网络预测部分增加一个极小目标预测头，有效增强多尺度、多目标的检测性能。较原YOLOv7算法，改进算法在自制数据集上mAP达到75.69%，提升了6.25%，对于多尺度特征的无人机多目标检测具有更好检测效果。 |
| 关键词: 无人机蜂群 光学检测 空中多目标检测 多尺度特征 特征融合 |
| DOI：10.20079/j.issn.1001-893x.230704002 |
|
| 基金项目: |
|
| A Multi-target Optical Detection Method for UAVBased on Improved YOLOv7 |
| ZHAO Qing,CHA Hao,MU Weiqi,LUO Yu |
| (1.College of Power Engineering,Naval University of Engineering,Wuhan 430033,China;2.Unit 32212 of PLA,Beijing 100036,China) |
| Abstract: |
| To solve the problems of slow detection speed and poor recognition in aerial multi-target detection with multi-scale features,an unmanned aerial vehicle(UAV) target detection algorithm based on improved YOLOv7 is proposed. For UAV swarm target,anchors of YOLOv7 are changed by using K-means algorithm to optimize the clustering of the self-built multi-scale drone dataset(MSDD).The network layers are extended in the neck network,so that the feature expression ability of small-scale targets is enhanced by further fusion of deeper features and superficial features.A minimal target prediction layer is added into the head network to improve the multi-scale and multi-target detection accuracy.Compared with that of the original YOLOv7 algorithm,the mAP of the improved algorithm reaches 75.69 % on the self-built dataset,which is improved by 6.25 %.The algorithm has better detection performance for UAV multi-target detection with multi-scale features. |
| Key words: UAV swarm optical detection aerial multi-target detection multi-scale features feature fusion |
《电讯技术》编辑部 