Answers are grounded in this paper and cite their sources.
Bird intrusions into airport security zones pose significant threats to aviation safety. However, the small size, rapid movement, and irregular flight patterns of birds often result in missed detections and false alarms in vision-based surveillance systems, especially in complex open-sky environments with low contrast and significant background interference. Although YOLOv8s achieves strong real-time performance, it remains limited in fine-grained feature representation and small-object detection capability. To address these challenges, we propose an improved detection model termed RPDL-YOLO (Refined Perception with Dual-branch Detection and Localization YOLO). The proposed model introduces three key enhancements: (1) integrating RepViTBlock into the backbone to improve spatial feature representation; (2) removing the redundant P5 detection head and constructing a dual-branch high-resolution P2 detection structure to strengthen small-object perception; and (3) designing a Mix_IoU loss function that combines EIoU and GIoU to improve localization accuracy and training stability. Experimental results demonstrate that, compared with the YOLOv8s baseline, the proposed method achieves improvements of 7.6% in Precision, 5.4% in Recall, and 4.6% in mAP@50, while maintaining a real-time inference speed of 93.1 FPS. These results demonstrate that RPDL-YOLO significantly enhances detection performance for small and fast-moving bird targets while maintaining computational efficiency, making it suitable for practical airport bird-strike prevention.
No takes yet. Share an insight, caveat, or question.
Wen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: