OnTech HSI Synthetic 21

The dataset comprise eight 128-by-128, 224 channels hyperspectral images. The images were generated using nine endmembers: Adularia GDS57, Jarosite GDS99, Jarosite GDS101, Anorthite HS349.1B, Calcite WS272, Alunite GDS83, Howlite GDS155, Corrensite CorWa-1, and Fassaite HS118.3B. Endmember spectra was taken from USGS spectral library. Per pixel abundances are provided for each image. Test images snr20, snr30, snr40, and snr50 uses corrupted endmembers to achieve the specified signal-to-noise ratio target. The dataset does not exhibit spatial coherence; therefore, it is best suited to train and evaluate pixel-oriented hyperspectral pixel unmixing models.

You can find the dataset here.

Please cite the following publication if you use this dataset.

@ARTICLE{24-ieee-igrs-j,
  author={Mantripragada, Kiran and Qureshi, Faisal Z.},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Hyperspectral Pixel Unmixing with Latent Dirichlet Variational Autoencoder}, 
  year={2024},
  volume={62},
  number={},
  pages={13pp},
  doi={10.1109/TGRS.2024.3357589},
  ISSN={1558-0644},
  month={January},
  keywords = {hsi-unmixing},
  url_Paper = {pubs/24-ieee-tgrs-j.pdf}
}

Publications

We have used this dataset in preparation of the following works.

License

This work is licensed under a Creative Commons Attribution 4.0 International License.