A Deep Learning-Based Hybrid Detection Approach for Reducing Bird Collisions in Wind Turbines

Öz

It is estimated that approximately 300,000 birds die annually worldwide due to collisions with wind turbines. In this study, an AI-assisted detection and intervention system was developed to minimize such collisions and enable wind turbines to operate without disrupting ecological balance. A dataset of 100 bird images under varying weather conditions and distances was collected, and the YOLOv4-tiny model was trained for 2000 epochs, achieving a mean average precision (mAP) of 95.8%. The trained model was deployed on an NVIDIA Jetson Nano and used with OpenCV-based image processing for real-time bird detection. When a bird is detected, an Arduino-controlled motor system stops the turbine blades, while an HC-SR04 ultrasonic sensor measures the bird’s distance to enhance decision stability. Testing under different lighting and distance conditions showed the best performance at high light and 30 cm (18 detections), and the lowest performance under low light and 42 cm (7 detections).

Anahtar Kelimeler

• Wind turbine
• bird safety
• artificial intelligence
• YOLO
• autonomous systems

Rüzgâr Türbinlerinde Kuş Çarpışmalarının Azaltılmasına Yönelik Derin Öğrenme Tabanlı Hibrit Algılama Yaklaşımı

Öz

Günümüzde rüzgar türbinleri nedeniyle yılda yaklaşık 300.000 kuşun yaşamını yitirdiği tahmin edilmektedir. Bu çalışmada, türbinlerin ekosisteme zarar vermeden çalışmasını sağlamak amacıyla yapay zekâ destekli bir kuş tespit ve müdahale sistemi tasarlanmıştır. Web üzerinden farklı hava koşulları ve uzaklıkları içeren 100 kuş görüntüsünden veri seti oluşturulmuş; YOLOv4-tiny modeli bu veriyle 2000 epok eğitilerek %95,8 mAP başarımı elde edilmiştir. Eğitilmiş model NVIDIA Jetson Nano üzerinde gerçek zamanlı olarak çalıştırılmış ve OpenCV tabanlı görüntü işleme ile kuş tespiti sağlanmıştır. Tespit durumunda Arduino kontrollü motor sistemi türbin pervanelerini durdurmakta, HC-SR04 sensörü ise kuş-türbin mesafesini ölçerek karar mekanizmasını desteklemektedir. Sistem, yüksek ışıkta ve 30 cm mesafede en iyi sonuçları (18 tespit), düşük ışıkta ve 42 cm mesafede ise en düşük performansı (7 tespit) göstermiştir.

Anahtar Kelimeler

• Rüzgar türbini
• kuş güvenliği
• görüntü işleme
• otonom sistemler
• çevresel sürdürülebilirlik

Kaynakça

1. Arnett, E. B., Huso, M. M. P., Schirmacher, M. R., & Hayes, J. P. (2011). Altering turbine speed reduces bat mortality at wind-energy facilities, Frontiers in Ecology and the Environment, 9(4), 209–214. https://doi.org/10.1890/100103
2. Ballester, P., May, R., & Reusch, C. (2024). A standardized protocol for evaluating automated detection systems to reduce wildlife collisions at wind farms, Journal of Environmental Management, 352, 119723. https://doi.org/10.1016/j.jenvman.2024.119723
3. De Lucas, M., Ferrer, M., Bechard, M. J., & Muñoz, A. R. (2020). Griffon vulture mortality at wind farms in southern Spain: Distribution of fatalities and active mitigation measures, Biodiversity and Conservation, 29(2), 397–414. https://doi.org/10.1007/s10531-019-01895-0
4. Desholm, M., & Kahlert, J. (2005). Avian collision risk at an offshore wind farm, Biological Conservation, 131(3), 312–322. https://doi.org/10.1016/j.biocon.2006.04.006
5. Happ, J., Gorresen, P. M., & Cryan, P. M. (2021). Automated daytime and nighttime monitoring of wildlife activity at wind turbines using visual and thermal cameras, Ecological Informatics, 64, 101367. https://doi.org/10.1016/j.ecoinf.2021.101367
6. Kekkonen, J., et al. (2025). Biologically inspired warning patterns: Adding red, black, and yellow to wind turbine blades could reduce bird collisions (preprint), bioRxiv. https://doi.org/10.1101/2025.04.14.648692
7. Klar, N., Gifary, N., Ziegler, F. P. G., et al. (2025). BirdRecorder’s AI on Sky: Detection and classification of endangered birds around wind turbines. arXiv preprint, arXiv:2508.18136*. https://arxiv.org/abs/2508.18136
8. Liu, Y., Zhang, H., & Chen, X. (2024). Bird detection near wind turbines using YOLOv4 in grayscale video, Sensors, 24(7), 2885. https://doi.org/10.3390/s24072885

9. May, R., Nygård, T., Falkdalen, U., Åström, J., Hamre, Ø., & Stokke, B. G. (2020). Paint it black: Efficacy of increased wind turbine rotor blade visibility to reduce avian fatalities. Ecology and Evolution, 10(16), 8927–8935. https://doi.org/10.1002/ece3.6592
10. McClure, C. J. W., Martinson, L., & Allison, T. D. (2021). Automated curtailment programs reduce eagle fatalities at wind energy facilities, Journal of Applied Ecology, 58(4), 858–868. https://doi.org/10.1111/1365-2664.13804
11. Oregon State University. (2024, August 20). Scientists studying impact of painting wind turbine blade black to reduce bird collisions. OSU Newsroom. https://news.oregonstate.edu/news/scientists-studying-impact-painting-wind-turbine-blade-black-reduce-bird-collisions
12. Pallanich, J. (2025, June). AI‑Driven bird monitoring reduces wind turbine collision risk. Journal of Petroleum Technology. https://jpt.spe.org/avian-intelligence-ai-driven-bird-monitoring-reduces-wind-turbine-collision-risk
13. Vattenfall. (2025, January). AI sheds light on bird collisions at offshore wind farms. Press release. https://group.vattenfall.com/press-and-media/newsroom/2025/ai-sheds-light-on-bird-collisions-at-offshore-wind-farms
14. Yarbrough, L., Farnsworth, A., & Sheldon, D. (2023). Automated classification of animals in thermal videos at wind energy facilities using deep learning (preprint), bioRxiv. https://doi.org/10.1101/2023.05.20.541556.