Region Based CNN for Foreign Object Debris Detection on Airfield Pavement

Area of Science:

• Computer Vision
• Artificial Intelligence
• Aerospace Engineering

Background:

• Foreign object debris (FOD) poses a significant safety risk in aviation.
• Current FOD detection methods often lack efficiency and accuracy.
• Automated detection systems are crucial for enhancing airport safety.

Purpose of the Study:

• To propose a novel algorithm for foreign object debris (FOD) detection using optical imaging sensors.
• To enhance the accuracy and efficiency of FOD detection systems.
• To improve upon existing deep learning-based object detection frameworks.

Main Methods:

• Development of a novel algorithm integrating an improved Region Proposal Network (RPN) and a Spatial Transformer Network (STN) based CNN classifier.
• Implementation of extra selection rules in the RPN to generate high-quality, fewer candidate regions.
• Integration of STN layer to enhance the efficiency of the CNN detector.

Main Results:

• The proposed algorithm demonstrated superior performance in detecting foreign object debris (FOD) on airfield pavement compared to Faster R-CNN and Single Shot MultiBox Detector (SSD).
• The improved RPN successfully generated fewer, higher-quality candidate regions.
• The inclusion of the STN layer significantly boosted the detection efficiency.

Conclusions:

• The novel CNN-based algorithm offers a promising solution for accurate and efficient FOD detection.
• The enhanced RPN and STN integration represent a significant advancement in FOD detection technology.
• This method has the potential to improve aviation safety by minimizing FOD-related risks.