Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Object-based urban land cover classification using YOLOv11s-seg on high-resolution orthophotos across municipalities in Scania, Sweden

Zvirblyté, Martyna LU (2026) NGEK11 20261
Department of Earth and Environmental Sciences (MGeo)
Abstract
This study evaluates an object-based image analysis (OBIA) approach using YOLOv11s-seg instance
segmentation model for classifying thermally relevant land cover types from 0.16 m resolution orthophotos across three municipalities in Scania, Sweden. Six thermally relevant classes were defined: asphalt, bright roof, brown roof, dark roof, tree canopy and grass. The model was trained on annotations from a single 2.5 × 2.5 km orthophoto tile in northern Lund and applied to other municipalities to assess transferability. The accuracy was assessed through random stratified sampling and visual interpretation of orthophotos. The model achieved a high overall accuracy of 91% in Lund, 90.6% in Svedala and 87.5% in Sjöbo as well as F1 scores of... (More)
This study evaluates an object-based image analysis (OBIA) approach using YOLOv11s-seg instance
segmentation model for classifying thermally relevant land cover types from 0.16 m resolution orthophotos across three municipalities in Scania, Sweden. Six thermally relevant classes were defined: asphalt, bright roof, brown roof, dark roof, tree canopy and grass. The model was trained on annotations from a single 2.5 × 2.5 km orthophoto tile in northern Lund and applied to other municipalities to assess transferability. The accuracy was assessed through random stratified sampling and visual interpretation of orthophotos. The model achieved a high overall accuracy of 91% in Lund, 90.6% in Svedala and 87.5% in Sjöbo as well as F1 scores of 0.910, 0.902 and 0.857 respectively. A distance decay pattern was observed in detection coverage as well as confidence, with less coverage and lower confidence further away from the training tile. Although precision and recall revealed a more complex pattern, tree canopy and roof classes were detected reliably across all municipalities while non-discrete surfaces such as asphalt and grass consistently struggled. The results indicate that YOLOv11s-seg can detect thermally relevant land cover in complex urban environments from high resolution orthophotos, while highlighting the architectural mismatch for non-discrete classes in the OBIA approach for land cover mapping. (Less)
Please use this url to cite or link to this publication:
author
Zvirblyté, Martyna LU
supervisor
organization
course
NGEK11 20261
year
type
M2 - Bachelor Degree
subject
keywords
YOLOv11, LULC, Scania, Sweden, OBIA, Instance segmentation, Spatial transferability, orthophoto, UHI
language
English
id
9238035
date added to LUP
2026-06-16 09:31:04
date last changed
2026-06-16 09:31:04
@misc{9238035,
  abstract     = {{This study evaluates an object-based image analysis (OBIA) approach using YOLOv11s-seg instance
segmentation model for classifying thermally relevant land cover types from 0.16 m resolution orthophotos across three municipalities in Scania, Sweden. Six thermally relevant classes were defined: asphalt, bright roof, brown roof, dark roof, tree canopy and grass. The model was trained on annotations from a single 2.5 × 2.5 km orthophoto tile in northern Lund and applied to other municipalities to assess transferability. The accuracy was assessed through random stratified sampling and visual interpretation of orthophotos. The model achieved a high overall accuracy of 91% in Lund, 90.6% in Svedala and 87.5% in Sjöbo as well as F1 scores of 0.910, 0.902 and 0.857 respectively. A distance decay pattern was observed in detection coverage as well as confidence, with less coverage and lower confidence further away from the training tile. Although precision and recall revealed a more complex pattern, tree canopy and roof classes were detected reliably across all municipalities while non-discrete surfaces such as asphalt and grass consistently struggled. The results indicate that YOLOv11s-seg can detect thermally relevant land cover in complex urban environments from high resolution orthophotos, while highlighting the architectural mismatch for non-discrete classes in the OBIA approach for land cover mapping.}},
  author       = {{Zvirblyté, Martyna}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Object-based urban land cover classification using YOLOv11s-seg on high-resolution orthophotos across municipalities in Scania, Sweden}},
  year         = {{2026}},
}