Optimizing Foreign Object Debris Detection on Runways with YOLO

With the rapid growth of the global aviation industry, foreign object debris (FOD) on airport runways poses a significant threat to flight safety, causing potential aircraft damage and severe accidents with casualties and economic losses. This study leverages deep learning, specifically YOLO series algorithms, to evaluate model performance for enhanced runway safety. Using a new dataset derived from FOD-A (3100 images, 31 debris types), we ensured data diversity through preprocessing. We tested nano and small versions of YOLOv5, YOLOv8, and YOLOv11 on 620 images, assessing accuracy, recall, F1-score, mAP50, mAP50-95, FPS, and model size. YOLOv8s excelled with an F1-score of 97.58%, mAP50 of 98.79%, and mAP50-95 of 89.56%, balancing accuracy and recall. YOLOv8n achieved 1460.43 FPS, ideal for real-time detection, while YOLOv11n led in mAP50-95 (90.12%), excelling in small debris detection. For resource-constrained scenarios, YOLOv5n and YOLOv8n are optimal due to their smaller size and speed, while YOLOv8s and YOLOv11n suit high-accuracy needs. YOLOv8n shows the best overall adaptability. This study validates YOLO’s efficiency in FOD detection, offering insights for aviation safety applications and future advancements in complex environments.