Enhancing road extraction accuracy with aerial LiDAR products and deep learning
Nordic Machine Intelligence · 17.06.2026 · 5 (2) · 7-16
DOI: 10.5617/nmi.12134 ↗Kurz gefasst
Fachartikel in der Zeitschrift Nordic Machine Intelligence, online erschienen am 17.06.2026. Verfasst von Jere Raninen und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Accurate road information is essential for urban planning, map updating, autonomous driving, and traffic management. Manual extraction is labor-intensive and error-prone, while aerial imagery suffers from occlusions such as shadows and vegetation. This study integrates LiDAR-derived products with RGB aerial imagery in a deep learning framework, leveraging LiDAR’s 3D structural information to overcome the limitations of optical data and improve road extraction accuracy. In this paper, we utilized the datasets including RGB imagery and LiDAR-derived products: aspect, DTM, hillshade, intensity, roughness, and slope, covering 23 urban and rural areas (864 km²). Two models, RoadVecNet and UNet 3+, were trained for 100 epochs. Performance was evaluated using the F1-score, precision, and recall. Results demonstrate that LiDAR features improved road extraction accuracy, with slope and roughness performing best. Hillshade and aspect contributed moderately but still enhanced performance. These findings highlight the effectiveness of combining aerial imagery and LiDAR for reliable road mapping in diverse environments.
Zitieren
APA
Raninen, J., Zhu, L., & Hattula, E. (2026). Enhancing road extraction accuracy with aerial LiDAR products and deep learning. Nordic Machine Intelligence, 5(2), 7-16. https://doi.org/10.5617/nmi.12134
Vancouver
Raninen J, Zhu L, Hattula E. Enhancing road extraction accuracy with aerial LiDAR products and deep learning. Nordic Machine Intelligence. 2026;5(2):7-16. doi:10.5617/nmi.12134
BibTeX
@article{Raninen2026,
author = {Raninen, Jere and Zhu, Lingli and Hattula, Emilia},
title = {Enhancing road extraction accuracy with aerial LiDAR products and deep learning},
journal = {Nordic Machine Intelligence},
year = {2026},
volume = {5},
number = {2},
pages = {7--16},
doi = {10.5617/nmi.12134}
}
EUVIMED