Models / hbku

hbku/satellite-landuse-classifier

Satellite Land-Use Classifier

PublicMinimal riskOpenComputer VisionPyTorchN/A1.2.0
2.1K 9.8K/30d 105

mIoU

0.81

Pixel accuracy

0.92

F1 (urban)

0.90

F1 (mangrove)

0.78

About this model

Semantic segmentation model that classifies Sentinel-2 imagery of Qatar into nine land-use classes including urban fabric, industrial, sabkha, mangrove, and reclaimed land. Produces 10 m resolution masks with per-pixel confidence suitable for change detection.

Intended use

Environmental monitoring, urban-expansion tracking, and coastal-zone reporting by planning and environment entities and researchers.

Training data lineage

Trained on qatar-sentinel2-tiles with 11,400 hand-annotated tile masks spanning 2021-2025 seasonal composites.

Limitations & bias notes

Mangrove and intertidal classes are confused at extreme low tide, and freshly graded construction sites are sometimes labelled as sabkha. Cloud and dust masking relies on upstream scene flags and can leak haze artefacts into predictions.

Evaluation metrics

mIoU0.81
Pixel accuracy0.92
F1 (urban)0.90
F1 (mangrove)0.78
#remote-sensing#sentinel-2#segmentation#land-use#environment

Try it

Live sandbox

sentinel2_alrayyan_2026Q2.tif

fixture

Feedback

Owning entity

hbHBKU
Updated
2026-04-19
Latest version
1.2.0
License
Open
Access
Open