Scholarlink

<p>With the increasing prevalence of small drones, the incidence of unauthorized and reckless drone operations—often referred to as “black flying” and indiscriminate flying—has become a growing concern, necessitating enhanced regulatory measures. However, existing UAV detection methods struggle to simultaneously balance detection accuracy, processing speed, and model complexity. To address these challenges, this paper introduces CD-YOLOv8s, a real-time, high-altitude UAV recognition model optimized for deployment on resource-constrained devices. CD-YOLOv8s achieves an optimal trade-off between accuracy and inference speed through several key innovations. First, the RFMDCSonv is introduced as a plug-and-play module to address the issue of convolutional kernel parameter sharing while dynamically capturing spatial and channel-wise relationships; secondly, the CCSM is incorporated to supplement coordinate information and refine feature extraction; and thirdly, the We_Concat operation is implemented in place of the conventional concatenation method, enabling improved feature fusion. To evaluate the proposed approach, a series of experiments is conducted on a publicly available drone dataset and the PASCAL VOC07+12 dataset. The proposed model achieves 77.12% mAP, 98.32% mAP50, 87.13% mAP75, and 121.3 FPS on the drone dataset, as well as 86.26% mAP50 on the VOC07+12 dataset. The results demonstrate that CD-YOLOv8s effectively balances accuracy, speed, and model complexity, making it well-suited for real-time UAV detection applications.</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

CD-YOLOv8s: an optimized high-altitude real-time UAV recognition method based on image detection

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Bing Su,
  • Jie Zhang,
  • Yifeng Lin

摘要

With the increasing prevalence of small drones, the incidence of unauthorized and reckless drone operations—often referred to as “black flying” and indiscriminate flying—has become a growing concern, necessitating enhanced regulatory measures. However, existing UAV detection methods struggle to simultaneously balance detection accuracy, processing speed, and model complexity. To address these challenges, this paper introduces CD-YOLOv8s, a real-time, high-altitude UAV recognition model optimized for deployment on resource-constrained devices. CD-YOLOv8s achieves an optimal trade-off between accuracy and inference speed through several key innovations. First, the RFMDCSonv is introduced as a plug-and-play module to address the issue of convolutional kernel parameter sharing while dynamically capturing spatial and channel-wise relationships; secondly, the CCSM is incorporated to supplement coordinate information and refine feature extraction; and thirdly, the We_Concat operation is implemented in place of the conventional concatenation method, enabling improved feature fusion. To evaluate the proposed approach, a series of experiments is conducted on a publicly available drone dataset and the PASCAL VOC07+12 dataset. The proposed model achieves 77.12% mAP, 98.32% mAP50, 87.13% mAP75, and 121.3 FPS on the drone dataset, as well as 86.26% mAP50 on the VOC07+12 dataset. The results demonstrate that CD-YOLOv8s effectively balances accuracy, speed, and model complexity, making it well-suited for real-time UAV detection applications.