Scholarlink

To address the challenges in the task of high-altitude cable detection using drones, such as the slender shape of targets, large scale variations, and complex background interference, this paper proposes a lightweight object detection algorithm with multi-module collaborative optimization based on improved YOLOv8obb. Firstly, the DCNv4 deformable convolution is introduced to construct the DE-SPPF module, enhancing the backbone network’s ability to capture linear features. Secondly, the MFC (Multi-scale Feature Cross Module) is designed, which enhances the model’s detection capability for slender targets of different scales and directions through multi-scale feature crossing and attention mechanisms, thereby improving robustness in complex backgrounds. Finally, the KLD loss function is combined to improve the localization accuracy of rotated targets. The improved model is named HC-YOLO. Experiments show that on the self-built high-altitude cable dataset HCS-2025, the mAP@0.5 of the proposed algorithm reaches 87.2%, which is 4.9% higher than that of the baseline model. It achieves high detection performance with a small increase in the number of model parameters, well balancing the model’s efficiency and performance.

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

Research on UAV-Based Object Detection Algorithm for High-Altitude Steel Cables

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yufeng Sun,
  • Liying Yang,
  • Xiaotian Yin,
  • Hua Xing

摘要

To address the challenges in the task of high-altitude cable detection using drones, such as the slender shape of targets, large scale variations, and complex background interference, this paper proposes a lightweight object detection algorithm with multi-module collaborative optimization based on improved YOLOv8obb. Firstly, the DCNv4 deformable convolution is introduced to construct the DE-SPPF module, enhancing the backbone network’s ability to capture linear features. Secondly, the MFC (Multi-scale Feature Cross Module) is designed, which enhances the model’s detection capability for slender targets of different scales and directions through multi-scale feature crossing and attention mechanisms, thereby improving robustness in complex backgrounds. Finally, the KLD loss function is combined to improve the localization accuracy of rotated targets. The improved model is named HC-YOLO. Experiments show that on the self-built high-altitude cable dataset HCS-2025, the mAP@0.5 of the proposed algorithm reaches 87.2%, which is 4.9% higher than that of the baseline model. It achieves high detection performance with a small increase in the number of model parameters, well balancing the model’s efficiency and performance.