A Comparative Evaluation of YOLO Architectures for Pothole Detection Using Multi-National Road Damage Datasets
Journal of Applied Science, Engineering and Technology for Development · 16.11.2026 · 6 (1)
DOI: 10.33803/rz2hvp.2 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Applied Science, Engineering and Technology for Development, online erschienen am 16.11.2026. Verfasst von Ephraim Mwangi Kinyua und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Potholes represent a persistent road safety and infrastructure maintenance challenge, particularly in regions where manual inspection is costly, slow, and geographically limited. This paper presents a comparative evaluation of deep learning-based pothole detection using multiple YOLO object detection architectures. A locally collected Kenyan pothole dataset (Chepalungu) is augmented with pothole-labeled images extracted from the multinational RDD2020 road damage dataset, spanning Japan, India, and the Czech Republic, in order to improve model generalization across varying road surfaces, lighting conditions, and camera viewpoints. Only the pothole annotation class is retained from RDD2020, and the merged dataset is subjected to integrity verification, negative-sample subsampling, and bounding-box statistical analysis prior to training. YOLOv8s is first evaluated at three input resolutions (640, 800, 960 pixels) to establish a suitable operating resolution. Then, the three YOLO architectures (YOLOv8s, YOLO11s, and YOLO26s) are trained and compared against each other at that resolution under identical training conditions. Model performance is assessed using precision, recall, mAP50, mAP50-95, and inference time. Experimental results indicate that increasing input resolution from 640 to 960 pixels improves detection performance, and that YOLO26s achieves the highest detection accuracy among the compared architectures while achieving inference speed comparable to YOLO11s and faster than YOLOv8s. These findings offer practical guidance for selecting YOLO configurations for efficient, resource-constrained pothole detection deployments in low-infrastructure road networks.
Zitieren
APA
Kinyua, E. M., Bwalei, M., Kihia, A., & Muguro, J. (2026). A Comparative Evaluation of YOLO Architectures for Pothole Detection Using Multi-National Road Damage Datasets. Journal of Applied Science, Engineering and Technology for Development, 6(1). https://doi.org/10.33803/rz2hvp.2
Vancouver
Kinyua EM, Bwalei M, Kihia A, Muguro J. A Comparative Evaluation of YOLO Architectures for Pothole Detection Using Multi-National Road Damage Datasets. Journal of Applied Science, Engineering and Technology for Development. 2026;6(1). doi:10.33803/rz2hvp.2
BibTeX
@article{Kinyua2026,
author = {Kinyua, Ephraim Mwangi and Bwalei, Macrey and Kihia, Alvin and Muguro, Joseph},
title = {A Comparative Evaluation of YOLO Architectures for Pothole Detection Using Multi-National Road Damage Datasets},
journal = {Journal of Applied Science, Engineering and Technology for Development},
year = {2026},
volume = {6},
number = {1},
doi = {10.33803/rz2hvp.2}
}
EUVIMED