Automatic Detection of Foreign Object Debris on Runways

Abstract

Foreign object debris can easily damage aircraft engines as well as injure personnel in an airport environment. Airfield Inspectors routinely inspect runways for the presence of FOD items, which differ in material, shape and colour using conventional and automated methods. The major shortcoming of the current method is their inability to detect all types of foreign objects in an accurate and timely manner. For removal from the airport runways in this study, we address this shortcoming, that is the lack of accuracy and timelines in the detection by developing an object detection framework to detect FOD for quick removal from the airports. The proposed FOD detection framework consists of unmanned aerial system for inspecting and collecting data from the airfields data processing and augmentation technique to counter the issue of learning on limited types of foreign objects, whether conditions and airport surface materials that are present in the data sets. A computer vision based object detection model to attain high accuracy and faster interference time force range and develop various models, including the you only look once object detector family of models in this framework.

Introduction

Foreign Object Debris (FOD) poses a major threat to aviation safety, often causing aircraft damage and accidents. Manual FOD inspection is inefficient, error-prone, and resource-intensive. This work proposes an automated FOD detection system using deep learning and computer vision to improve accuracy, efficiency, and safety in real-time runway monitoring.

Key Components of the System:

• Hardware Setup: Drones equipped with RGB cameras fly over runways (~30m altitude), scanning for debris.

• Onboard Processing: Real-time detection using YOLOX and CNN-based classifiers distinguishes between metal and non-metal debris.

• Navigation: Autonomous path planning, obstacle avoidance, and precision hovering over detected FOD.

• FOD Removal: Mechanism includes grippers for collection or GPS-beacon placement for manual retrieval.

• Communication & Reporting: Detection data is wirelessly transmitted for real-time monitoring and logged for analysis.

Software Flow:

1. Capture image.

2. Detect object patterns.

3. Extract features (size, shape, color).

4. Match features with trained database.

5. Identify and classify FOD.

Hybrid Algorithm Flow:

• Train CNN with global pooling.

• Extract GAP features and train SVM classifier.

• Evaluate on test data for accuracy.

Dataset & Training:

• 4000 images across 3 classes (metal parts, stones, other FOD like plastic/wood).

• Images sourced from Kaggle and airport footage.

• Model trained on input image sizes up to 4992x3328.

Results:

• Achieved 80%+ detection accuracy, even in blurred or poor visibility conditions.

• CNN model shows high confidence in identifying nuts, bolts, and stones.

• Demonstrated reliable real-time classification and detection under diverse airport conditions.

Conclusion

The presence of a large number of foreign objects (in various colors, shapes, and materials) in an airport environment makes it infeasible to train an object detection model on all types of FOD items that could be found on runways. This characteristic makes the FOD items a set of unbounded target objects and this problem non-trivial. The data augmentation component in the proposed FOD detection framework is able to counter the issue of training on limited types of foreign objects, weather conditions, and surface materials and attain higher FOD detection accuracy. It is important to detect foreign objects, irrespective of identifying their types, to reduce the chance of damage or injury in an airport environment. By assigning a single class to the set of foreign objects in the training data, we demonstrated that our framework with the computer vision model is able to learn the inherent features within the internal filters and detect previously unseen objects with a high accuracy. The CNN model trained on the data set of FOD images obtained after applying the augmentation techniques on the original images captured by the UAS outperforms all the other modeling approaches in the object detection accuracy performance metrics. The proposed FOD detection framework with the CNN model has the highest average precision value, recall rate, and precision rate among all other models.

References

[1] Qingqing Li, Xiaojun Tan, and Shuai Wang “Research on FOD Detection System of Airport Runway Based on Artificial Intelligence” 2022. [2] Xiaolin Qin, Xiaoyi Wang, and Xianjun Qi “Airport Runway Foreign Object Debris Detection Based on YOLOX Architecture” 2023. [3] Shufan Yang, Xiaolin Qin, and Wei Wang “Foreign Object Debris Automatic Target Detection for Millimeter-Wave Surveillance Radar” 2021. [4] Wen-Jing Wang and Yun-Fei Wang “Airport Runway Foreign Object Debris Detection System Based on Arc-Scanning SAR Technology” 2021. [5] Shufan Yang, Xiaolin Qin, and Lei Wang “Foreign Object Debris Classification Through Material Recognition Using Deep Convolutional Neural Network with Focus on Metal” 2021. [6] Xianjun Qi, Xiaolin Qin, and Wei Wang “Foreign Object Debris Detection on Airport Runways”: A Comprehensive Survey (2020).

Copyright

Copyright © 2025 Dr. M. Rajeswari, Prakruthi D S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.