Performance Benchmarking of Lightweight YOLO Architectures for Drone Detection in Thermal Imagery
11th International Conference on Recent Advances in Air and Space Technologies, Conference Program, RAST 2026, İstanbul, Türkiye, 13 - 15 Mayıs 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/rast69551.2026.11672391
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: autonomous surveillance, counter-UAV Systems, edge computing, real-time detection, thermal computer vision, YOLO
- Ankara Üniversitesi Adresli: Evet
Özet
Nowadays, with the uncontrolled widespread use of unmanned aerial vehicles (UAVs), these vehicles have become a security risk and a threat in both civilian and defense airspace. For this reason, there is a need for continuous and autonomous surveillance systems. Thermal imaging offers a passive detection method that is of critical importance in low-contrast and longrange scenarios, night operations, and challenging conditions. While these advantages are strategically effective, the large volume of data limits real-time drone detection capabilities on edge platforms with limited hardware, creating a bottleneck that must be addressed. This study comparatively analyzed drone detection performance in the thermal spectrum using the latest lightweight YOLO (You Only Look Once) architectures: the YOLOv8n, YOLOv11n, and YOLOv12n models. Detection performance was comprehensively compared in terms of accuracy and efficiency using a dataset consisting of 3,950 unique thermal drone images. The YOLOv12n model offers the most up-to-date and reliable architecture for defense applications in the aviation sector, achieving an mAP@50 of 87% and a recall rate of 85%. YOLOv11n, with an inference time of 2.91 ms and a frame rate of 343.46 FPS, reduces computational costs and, due to its lightweight and fast nature, provides unmatched performance advantages in autonomous intervention systems. The findings demonstrate that the choice of technological architecture in next-generation UAV security and air surveillance can fundamentally shift the balance of these fields.