Zurich Summer: 20 QuickBird Satellite Chips with Urban Land Cover Labels

20 high-resolution satellite image chips from Zurich, Switzerland, manually annotated with 8 urban land cover classes. The dataset was created by researchers at the University of Edinburgh from a QuickBird acquisition in August 2002. Annotations include roads, buildings, trees, grass, bare soil, water, railways, and swimming pools, with an intentionally unbalanced class distribution.

Use Cases

• Training semantic segmentation models based on the 8 annotated urban land cover classes.
• Benchmarking remote sensing classification algorithms based on high-resolution pansharpened QuickBird imagery.
• Studying class imbalance in real-world urban scenes based on the described unbalanced sample distribution.
• Testing model generalization on a small, controlled set of georeferenced image tiles.

Strengths

• Provides 20 georeferenced image-ground truth pairs in GeoTIFF format.
• Includes 8 manually annotated urban and peri-urban land cover classes.
• Class distribution is intentionally unbalanced to reflect real-world situations.
• Comes with a MATLAB preprocessing script for normalization, visualization, and format conversion.

Limitations

• Dataset scale is small, consisting of only 20 image chips.
• Annotations are described as imperfect, not ultradense, and may contain errors.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

University of Edinburgh researchers Vittorio Ferrari and Michele Volpi, using DigitalGlobe QuickBird imagery.

Collection Method

Manual annotation via superpixel selection and freehand drawing on pansharpened satellite imagery.

Time Range

August 2002

Geography

City of Zurich, Switzerland