Transfer Learning-based Approach for Off-Road Route Identification in the Realm of Autonomous Terrain Analysis and Decision Making
Defence Science Journal · 22.06.2026 · 76 (4) · 511-522
DOI: 10.14429/dsj.21132 ↗Kurz gefasst
Fachartikel in der Zeitschrift Defence Science Journal, online erschienen am 22.06.2026. Verfasst von Manjeet Kaur und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 2.5).
Abstract (Originalsprache)
Autonomous traversability in unstructured roads demands automatic off-road track identification systems. This area has garnered significant research interest in recent times. The present study leverages the YOLOv8 transfer learning architecture and its variants for off-road track identification and segmentation to detect drivable tracks in off-road areas. The YOLOv8 model has been chosen for their advanced feature extraction capabilities. Currently, the off-road areas present the challenges of loosely organised class boundaries, non-uniform features, and a noisy environment. Thus, the present study provides an end-to-end strategy using an optimised YOLOv8l variant for terrain analysis and emergency decision-making, offering an alternative to conventional image processing and LiDAR-based approaches. The YOLOv8 model and its variants were evaluated and compared across both validation and inference datasets. Out of which the YOLOv8l model was tuned for selectable hyperparameters from the internal architecture to balance speed, computational efficiency, and detection precision across the large data set. The reliability of the optimised YOLOv8l model was evaluated on a real-time video feed and on unseen image scenes captured with a camera, and the model achieved an overall accuracy of more than 94 %. The model demonstrated its efficiency in snow regions for identifying unstructured tracks, showcasing its potential for autonomous military and emergency navigation operations in high-altitude conditions.
Zitieren
APA
Kaur, M., Mittal, U., Sharma, V., & Kalra, M. (2026). Transfer Learning-based Approach for Off-Road Route Identification in the Realm of Autonomous Terrain Analysis and Decision Making. Defence Science Journal, 76(4), 511-522. https://doi.org/10.14429/dsj.21132
Vancouver
Kaur M, Mittal U, Sharma V, Kalra M. Transfer Learning-based Approach for Off-Road Route Identification in the Realm of Autonomous Terrain Analysis and Decision Making. Defence Science Journal. 2026;76(4):511-522. doi:10.14429/dsj.21132
BibTeX
@article{Kaur2026,
author = {Kaur, Manjeet and Mittal, Upika and Sharma, Vikas and Kalra, M.K.},
title = {Transfer Learning-based Approach for Off-Road Route Identification in the Realm of Autonomous Terrain Analysis and Decision Making},
journal = {Defence Science Journal},
year = {2026},
volume = {76},
number = {4},
pages = {511--522},
doi = {10.14429/dsj.21132}
}
EUVIMED