Scholarlink

<p>The Foreign Object Debris (FOD) on airport runways often disrupts operations while taking off and landing flights, leading to accidents. Small-scale FOD cannot manually rule out on time, which threatens aviation safety. To address these issues, this article proposes an improved real-time detection model, SPLP-YOLO, based on the You Only Look Once (YOLO) architecture, which integrates multiple innovative designs for efficient and accurate FOD detection. Firstly, a spatial-to-depth (SPD) convolution module is introduced to preserve detailed information of small targets. Secondly, the novel Coarse-to-Fine Large Separable Kernel Attention (C2F-LSKA) module is proposed, combining multi-branch feature diversion with separable large-kernel attention mechanisms to enhance local and global feature modeling capabilities. Moreover, the Bi-level Routing Attention (BRA) mechanism is employed to refine attention modeling and improve small-object detection accuracy. Additionally, a Powerful-IoU (PIoU) loss function is designed to accelerate convergence and optimize bounding box regression. The model further enhances small-object detection by incorporating a fourth detection head. Experimental results demonsstrate superior performance of SPLP-YOLO on both self-constructed and open-source FOD-A datasets, achieving mAP@0.5:0.95 scores of 0.67 and 0.936, respectively, which represent improvements of over 10% compared to Baseline, while maintaining an inference time of just 4.2ms (240 Frames Per Second). The proposed model also exhibits robust performance in complex environments such as rain, fog, and low illumination, providing a lightweight, highly accurate real-time solution for airport runway safety monitoring.</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

Splp-yolo: an all-weather real-time detector for airport runway foreign object debris

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Runze Zhang,
  • Ce Wang,
  • Bingru Li,
  • Ouming Ye,
  • Xudong Xu

摘要

The Foreign Object Debris (FOD) on airport runways often disrupts operations while taking off and landing flights, leading to accidents. Small-scale FOD cannot manually rule out on time, which threatens aviation safety. To address these issues, this article proposes an improved real-time detection model, SPLP-YOLO, based on the You Only Look Once (YOLO) architecture, which integrates multiple innovative designs for efficient and accurate FOD detection. Firstly, a spatial-to-depth (SPD) convolution module is introduced to preserve detailed information of small targets. Secondly, the novel Coarse-to-Fine Large Separable Kernel Attention (C2F-LSKA) module is proposed, combining multi-branch feature diversion with separable large-kernel attention mechanisms to enhance local and global feature modeling capabilities. Moreover, the Bi-level Routing Attention (BRA) mechanism is employed to refine attention modeling and improve small-object detection accuracy. Additionally, a Powerful-IoU (PIoU) loss function is designed to accelerate convergence and optimize bounding box regression. The model further enhances small-object detection by incorporating a fourth detection head. Experimental results demonsstrate superior performance of SPLP-YOLO on both self-constructed and open-source FOD-A datasets, achieving mAP@0.5:0.95 scores of 0.67 and 0.936, respectively, which represent improvements of over 10% compared to Baseline, while maintaining an inference time of just 4.2ms (240 Frames Per Second). The proposed model also exhibits robust performance in complex environments such as rain, fog, and low illumination, providing a lightweight, highly accurate real-time solution for airport runway safety monitoring.