FOD detection research using BSM-YOLO during construction without air service suspension

Area of Science:

• Computer Vision
• Artificial Intelligence
• Aerospace Engineering

Background:

• Airport expansion necessitates Foreign Object Debris (FOD) detection during construction.
• Traditional FOD detection models struggle with complex construction environments and multi-scale targets.
• Current methods often require air service suspension, impacting operations.

Purpose of the Study:

• Introduce the BSM-YOLO model for improved FOD detection in complex airport construction scenarios.
• Enhance multi-scale target recognition and adaptability to image variations.
• Provide a solution for safe, continuous airport operations during construction.

Main Methods:

• Enhanced Bidirectional Feature Pyramid Network (BIFPN) with independent pyramid branches and rapid normalization for optimized feature fusion.
• Incorporated Sic2f structure with similarity attention in the backbone for improved adaptability to complex image variations.
• Introduced Mc2f structure at the neck level using multi-scale convolutions and channel attention for spatial-frequency information capture.

Main Results:

• The BSM-YOLO model achieved a 5.3% increase in mean Average Precision (mAP) compared to the YOLOv8n model.
• Demonstrated superior performance in multi-scale FOD detection within complex construction environments.
• Validated the model's effectiveness in scenarios requiring uninterrupted air service.

Conclusions:

• The BSM-YOLO model significantly improves FOD detection accuracy and robustness.
• Offers a viable solution for enhancing safety during airport construction without service suspension.
• Contributes to the advancement of AI-driven safety solutions in aviation infrastructure projects.