Scholarlink

<p>Foreign object debris (FOD) on airport runways poses a significant threat to aviation safety. Traditional detection methods, which rely on manual inspections and radar-based systems, struggle to meet the increasing demands of modern airport operations. Unmanned aerial vehicles (UAVs) equipped with optical sensors offer a promising solution by improving detection efficiency while reducing operational costs. This paper presents a novel FOD detection method specifically designed for UAV platforms, together with the construction of a dedicated dataset, UAV-FOD. The dataset comprehensively covers diverse environmental conditions, minute and multi-scale FOD instances, as well as varied viewing angles and flight altitudes. Consequently, it effectively captures the distinctive challenges of UAV-based FOD detection, particularly the need for high sensitivity to small objects and robust multi-scale feature extraction. To address the challenges of FOD recognition, a lightweight and efficient detection model, termed YOLO-URD, is proposed. Built upon YOLOv11, the model integrates an ADown module that employs a dual-branch collaborative compression architecture, significantly reducing model parameters while enhancing the extraction of fine-grained FOD features. In addition, a C3k2-ACG module is introduced, which leverages a multi-component guidance mechanism to emphasize critical feature regions, thereby improving sensitivity to multi-scale FOD targets. Furthermore, an LSQE detection head is designed to generate a novel classification score by jointly evaluating classification confidence and regression quality, which improves recognition accuracy for clustered FOD distributions while maintaining model efficiency. Moreover, a Wise-Inner-PIoUv2 loss function is developed by integrating advantageous components from multiple loss formulations to further refine localization precision. Experimental results demonstrate that YOLO-URD achieves 94.1% precision, 84.6% recall, 89.1% F1-score, and 90.2% mAP, with only 1.81 million parameters and 4.2 GFLOPs. Compared with the baseline model, YOLO-URD improves precision by 2.6%, recall by 4.0%, and mAP by 4.1%, while reducing the number of parameters by 29.8% and computational complexity by 33%. These results indicate that the proposed method provides an efficient and effective UAV-based solution for FOD detection, contributing to the advancement of real-time runway monitoring systems.</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

YOLO-URD: a lightweight UAV-based model for foreign object detection on airport runways

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月17日

查看全文
  • Biao Wang,
  • Qing Tao,
  • Wei Liu,
  • Fanhuan Li

摘要

Foreign object debris (FOD) on airport runways poses a significant threat to aviation safety. Traditional detection methods, which rely on manual inspections and radar-based systems, struggle to meet the increasing demands of modern airport operations. Unmanned aerial vehicles (UAVs) equipped with optical sensors offer a promising solution by improving detection efficiency while reducing operational costs. This paper presents a novel FOD detection method specifically designed for UAV platforms, together with the construction of a dedicated dataset, UAV-FOD. The dataset comprehensively covers diverse environmental conditions, minute and multi-scale FOD instances, as well as varied viewing angles and flight altitudes. Consequently, it effectively captures the distinctive challenges of UAV-based FOD detection, particularly the need for high sensitivity to small objects and robust multi-scale feature extraction. To address the challenges of FOD recognition, a lightweight and efficient detection model, termed YOLO-URD, is proposed. Built upon YOLOv11, the model integrates an ADown module that employs a dual-branch collaborative compression architecture, significantly reducing model parameters while enhancing the extraction of fine-grained FOD features. In addition, a C3k2-ACG module is introduced, which leverages a multi-component guidance mechanism to emphasize critical feature regions, thereby improving sensitivity to multi-scale FOD targets. Furthermore, an LSQE detection head is designed to generate a novel classification score by jointly evaluating classification confidence and regression quality, which improves recognition accuracy for clustered FOD distributions while maintaining model efficiency. Moreover, a Wise-Inner-PIoUv2 loss function is developed by integrating advantageous components from multiple loss formulations to further refine localization precision. Experimental results demonstrate that YOLO-URD achieves 94.1% precision, 84.6% recall, 89.1% F1-score, and 90.2% mAP, with only 1.81 million parameters and 4.2 GFLOPs. Compared with the baseline model, YOLO-URD improves precision by 2.6%, recall by 4.0%, and mAP by 4.1%, while reducing the number of parameters by 29.8% and computational complexity by 33%. These results indicate that the proposed method provides an efficient and effective UAV-based solution for FOD detection, contributing to the advancement of real-time runway monitoring systems.