Suitability of Latest Version of YOLOv11 in Drone Development Studies

Although drones have become increasingly advanced and sophisticated, they still face challenges such as limited computing power and high-power consumption. Optimizing their features is a key strategy to address these limitations. One critical feature that can be further improved is the algorithm for specialized tasks like obstacle detection. Recent studies have primarily relied on earlier versions of YOLO for this purpose. However, with the release of the latest YOLO versions, YOLOv10 and YOLOv11, there is an opportunity to assess their suitability for drone applications. In this study, we trained YOLOv10-nano and YOLOv11-nano using a drone-specific dataset and compared their performance with similar studies. The results indicate that YOLOv11-nano outperforms YOLOv10-nano in bounding-box prediction, achieving a higher mAP of 0.957 with a 0.22 difference, as well as faster inference times with a 74.1 ms improvement, resulting in a frame rate gain of 0.0421 FPS. However, when compared to earlier studies, YOLOv11-nano demonstrates superior precision but lags in real-time processing speed. These findings highlight the trade-off between accuracy and inference time, highlighting the suitability of YOLOv11-nano for applications in drone development where accuracy takes precedence over processing speed.