Benchmarking YOLO Variants for Vision-Based Flap State Detection in Fixed-Wing UAVs
International Research Journal of Multidisciplinary Technovation · 26.05.2026 · 354-370
DOI: 10.54392/irjmt26321 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Research Journal of Multidisciplinary Technovation, online erschienen am 26.05.2026. Verfasst von Srinivasan V und 1 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Fixed-wing UAVs depend on their flap state for safe operation, to do that conventional sensors cannot be used such as the potentiometers because they are rigid, weigh so much, and lack adaptation according to the place they are used. This creates a critical gap in the need for a good fault-tolerant method used for monitoring the UAV’s control surface. This study aims to benchmark this fault tolerant method, as the fault tolerant method used here will be using Oak D Lite stereo vision camera and the implementation of YOLO model in it, the YOLO model to be used is benchmarked for superior performance, thus establishing a good and lightweight alternative to traditional sensors. A thorough dataset of about 1,920 raw images were taken by mounting the Oak D Lite camera at the tail section of the fixed-wing UAV, it has six flap state images taken under various lighting conditions, depth, and background conditions. After augmentation, 4,623 images are been used to train and then evaluate the YOLOv5(s/m/l), YOLOv8(s/m/l), and YOLO11(s/m/l) models, other models are either too old or too much minimal for this research. These models are evaluated under standard parameters run on an RTX 4090 GPU. Found out that all the models achieved mAP@0.5 exceeding 98%, with a good precision and recall consistency that are above 98.1% and 95.5% respectively. YOLOv5s achieved the best F1-score with the lowest confidence threshold with only 7 misdetections, while other models like YOLOv5s and YOLOv8s had a good inference speed of about 0.6ms. These findings provide a validated baseline benchmark for UAV control surface monitoring and offer practical deployment guidelines for real-time fault-tolerant UAV systems.
Zitieren
APA
V, S., & J.V, S. P. K. (2026). Benchmarking YOLO Variants for Vision-Based Flap State Detection in Fixed-Wing UAVs. International Research Journal of Multidisciplinary Technovation, 354-370. https://doi.org/10.54392/irjmt26321
Vancouver
V S, J.V SPK. Benchmarking YOLO Variants for Vision-Based Flap State Detection in Fixed-Wing UAVs. International Research Journal of Multidisciplinary Technovation. 2026:354-370. doi:10.54392/irjmt26321
BibTeX
@article{V2026,
author = {V, Srinivasan and J.V, Sai Prasanna Kumar},
title = {Benchmarking YOLO Variants for Vision-Based Flap State Detection in Fixed-Wing UAVs},
journal = {International Research Journal of Multidisciplinary Technovation},
year = {2026},
pages = {354--370},
doi = {10.54392/irjmt26321}
}
EUVIMED