LandCover.CARTOSAT-2E: A High-Resolution Multispectral Dataset for Land Cover Semantic Segmentation

LandCover.CARTOSAT-2E is a high-resolution multispectral remote sensing dataset developed for semantic segmentation of land cover from satellite imagery. The dataset is derived from ISRO's CARTOSAT-2E satellite imagery acquired over Bhubaneswar, Odisha, India, and contains four spectral bands (RGB + Near-Infrared) together with manually annotated pixel-wise ground truth masks for six land cover classes.

The dataset has been developed to support research in semantic segmentation, land cover mapping, urban monitoring, and deep learning for remote sensing applications. Standard training, validation, and test splits are provided to facilitate reproducible benchmarking.

Dataset summary

• Satellite: CARTOSAT-2E
• Location: Bhubaneswar, Odisha, India
• Spatial resolution: 1.6 m
• Spectral bands: RGB + NIR (4-band)
• Annotation: Pixel-wise semantic segmentation
• Number of classes: 6
• Image format: GeoTIFF (.tif)
• Mask format: PNG (.png)

Comprehensive documentation, including class definitions, dataset organization, train/validation/test split, and licensing information, is provided in the accompanying dataset_description.pdf and README.md files.

Citation

If you use this dataset in your research, please cite both the dataset and the accompanying publication.

Dataset

Shrivastava, V., Bera, S., & Srivastava, V. (2026).
LandCover.CARTOSAT-2E: A High-Resolution Multispectral Dataset for Land Cover Semantic Segmentation.
Zenodo.
https://doi.org/10.5281/zenodo.21378148

Associated publication

Shrivastava, V. K., et al. GAFNet: Global-Attention Fusion Network for Land Cover Segmentation in Remote Sensing Imagery. (Manuscript under review; publication details will be updated upon acceptance.)