Earticle

초록

영어

In this study, I apply the Selected Query Detection Transformer, a powerful transformer-based framework for real-time bird detection. This research aims to improve the reliability of bird detection to address bird strikes, a serious threat to aviation safety. The proposed model was trained using the CUB-200-2011 dataset and subsequently fine-tuned on a real-world bird surveillance dataset collected near airport runways to enhance adaptability to real environments. SQ-DETR employs a layer-wise adaptive query pruning mechanism that dynamically removes low-importance object queries during decoding, thereby reducing redundant computations while preserving detection accuracy. Experimental results demonstrate that SQ-DETR outperforms YOLOv8-L, achieving 2.5% higher mean Average Precision and reducing computational cost by approximately 18%, with an AP₅₀ of 90.0 and AP₇₅ of 87.2. Qualitative analysis further shows that SQ-DETR more accurately detects small or partially occluded birds in complex airport scenes compared to YOLO. Overall, SQ-DETR effectively balances precision and efficiency, providing a practical and scalable framework for real-time bird surveillance and bird-strike prevention systems. This study highlights the potential of Transformer-based architectures to enhance safety and operational reliability in modern aviation monitoring environments.

목차

Abstract
1. Introduction
2. Related works
3. Methods
4. Experimental Results
4.1 Dataset and Fine-Tuning Setup
4.2 Quantitative Comparison
4.3 Effect of Decoder Depth
4.4 Qualitative Analysis
5. Discussion
6. Conclusion
Acknowledgement
References

키워드

저자

• Heon Jeong [ Professor, Department of Fire Administration, Chodang University, Korea ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  The International Journal of Advanced Smart Convergence

• 간기

  계간

• pISSN

  2288-2847

• eISSN

  2288-2855

• 수록기간

  2012~2025

• 십진분류

  KDC 326 DDC 380