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

As aviation activities expand, the frequency of airport expansion and renovation projects has increased, making the detection of Foreign Object Debris (FOD) during construction without air service suspension critically important. However, traditional FOD detection models have limitations in handling complex construction environments and multi-scale target recognition. To address these challenges, this paper introduces the innovative BSM-YOLO model. This model first enhances the Bidirectional Feature Pyramid Network(BIFPN)structure by incorporating two independent pyramid branches and implementing a rapid normalization method. These enhancements improve the multi-level fusion of feature maps and cross-level feature interactions, optimizing the performance of multi-scale target detection. Additionally, inspired by attention mechanisms, the Sic2f structure is innovatively designed within the backbone network. This structure uses a similarity attention module to capture correlations between modalities, Improving the model's adaptability to complex image variations. Finally, the Mc2f structure is introduced at the neck level. This structure effectively captures spatial and frequency information through multi-scale convolutions and channel-level attention mechanisms. The BSM-YOLO model demonstrates an improvement over the YOLOv8n model, with increases of 5.3% in mean Average Precision(mAP). These enhancements confirm the model's superiority in managing complex scenarios during construction without air service suspension and in multi-scale FOD detection, offering substantial benefits for improving safety operations during such periods.