Detection and classification of foreign object debris (FOD) with comparative deep learning algorithms in airport runways

Air transportation is one of the fastest and safest modes of transportation in our time. However, the safety and efficiency of air travel require effective management of various risk factors. One of these risks is foreign object damage (FOD) on airport runways. Since foreign object hazard can cause aircraft to receive critical damage during takeoff and landing, FOD detection is of great importance for air transportation security. In this study, main aim is to detect and classify FOD by employing deep learning method involving YOLOv5 (CSP-Darknet53 architecture) and YOLOv8 (Pytorch architecture) versions. In this context, a new dataset called FOD-Runway consisting of seventy-one different class objects that could be likely found in runway is obtained where database is enlarged with 33,286 images by data augmentation methods. Moreover, the obtained database is subjected to deep learning methods such as YOLO models and detection success of the models is quantified with recall, precision, F-measure and mAP. According to the analysis outcomes, F-measure is obtained the highest 0.894 by YOLOv5m ad 0.907 by YOLOv8x. Furthermore, mAP0.5 value is obtained as 0.911 by YOLOv5m and 0.939 by YOLOv8x whereas mAP0.5-0.95 value is computed as 0.868 and 0.939, respectively. It could be deduced that this study serves in boosting safety of airport thanks to the obtained dataset called FOD-Runway, which involves more FOD class than existing dataset. Due to this consideration, it could contribute in increasing precision to deep learning based FOD detection system.