Scholarlink

<p>It is difficult to achieve precise positioning of unmanned aerial vehicle (UAV) based on GPS signals in complex environments or indoors. The acquisition of UAV location relies on visual information rely on visual information. In order to effectively identify UAV in the formation and reduce detection costs, we proposed a lightweight recognition network based on YOLOv5, which was inspired by edge computing. The network was suitable for front-end deployment and aimed to solve the problem that YOLOv5s could not meet the running speed of edge detection. Specifically, an improved convolutional neural network based on YOLOv5s was applied by modifying the network structure and combining the Ghostbottleneck module of Coordattention. The experimental results showed that our improved YOLO algorithm achieved the speed by 9.5%, recall by 1.8% and precision by 1.7% compared with YOLOv5s. The transformed model was deployed on NVIDIA’s Jetson nano edge computing platform. Its running speed could reach 13 FPS, which could better meet the actual engineering needs of UAV detection in formation flight.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An improved YOLO algorithm for UAV detection in formation flight

  • 期刊论文

  • 发表日期：

    2025年1月17日

查看全文
  • Shuo Wang,
  • Yangyang Liu,
  • Xinhua Wang,
  • Jinfa Xu

摘要

It is difficult to achieve precise positioning of unmanned aerial vehicle (UAV) based on GPS signals in complex environments or indoors. The acquisition of UAV location relies on visual information rely on visual information. In order to effectively identify UAV in the formation and reduce detection costs, we proposed a lightweight recognition network based on YOLOv5, which was inspired by edge computing. The network was suitable for front-end deployment and aimed to solve the problem that YOLOv5s could not meet the running speed of edge detection. Specifically, an improved convolutional neural network based on YOLOv5s was applied by modifying the network structure and combining the Ghostbottleneck module of Coordattention. The experimental results showed that our improved YOLO algorithm achieved the speed by 9.5%, recall by 1.8% and precision by 1.7% compared with YOLOv5s. The transformed model was deployed on NVIDIA’s Jetson nano edge computing platform. Its running speed could reach 13 FPS, which could better meet the actual engineering needs of UAV detection in formation flight.