🛰️ HSI Datasets Collection

A comprehensive collection of 24 publicly available Hyperspectral Image (HSI) datasets curated for research in hyperspectral image classification, land use/land cover (LULC) mapping, and remote sensing deep learning benchmarks.

Total Size: ~20.1 GB
License: Apache 2.0
Maintained by: Tanishq Rachamalla, Aryan Das
HuggingFace Dataset: https://huggingface.co/datasets/Tanishq165/HSI_Datasets
Paper: Hyperspectral Image Models: Technical Report (arXiv:2609.39871)
Code: Hyperspectral Image Models — A PyTorch Library for Hyperspectral Image Models

---

📖 Dataset Description

Hyperspectral imaging captures data across hundreds of narrow spectral bands, enabling fine-grained material and land cover discrimination beyond what RGB or multispectral sensors can achieve. This repository consolidates 24 widely-used HSI benchmark datasets in a single place, covering a variety of sensors (AVIRIS, ROSIS, Hyperion, CRISM), geographic regions (USA, Europe, Asia, Mars), and scene types (urban, agricultural, coastal, planetary).

All datasets are stored in .mat format (MATLAB-compatible), making them directly loadable with scipy.io in Python.

---

📦 Included Datasets

All datasets are hosted on the HuggingFace Hub: Tanishq165/HSI_Datasets

---

🗂️ Repository Structure

HuggingFace Hub Structure

HSI_Datasets/
├── Augsburg/
├── Berlin/
├── Botswana/
├── Chikusei/
├── Dioni/
├── Holden/
├── Houston13/
├── Houston18/
├── Indian_Pines/
├── KSC/
├── Loukia/
├── Muufl/
├── NiliFossae/
├── Pavia_Center/
├── Pavia_University/
├── Pingan/
├── Qingyun/
├── Salinas/
├── Tangdoaowan/
├── Trento/
├── Utopia/
├── WHU-Hi-HanChuan/
├── WHU-Hi-HongHu/
└── WHU-Hi-LongKou/

Each dataset folder contains:

DatasetName/
├── data.mat        # Hyperspectral cube: (H × W × Bands)
└── gt.mat          # Ground truth label map: (H × W)

---

🚀 Loading the Data

Install dependencies

pip install huggingface_hub scipy numpy h5py scikit-learn pyyaml matplotlib

Load a single dataset (via project loader)

from utils.data_loader import DatasetLoader
import numpy as np

loader = DatasetLoader(use_cache=True)

# Auto-downloads from HuggingFace Hub on first use
hsi_cube, labels = loader.load_dataset("Indian_Pines")

print(f"HSI Cube shape : {hsi_cube.shape}")
print(f"Labels shape   : {labels.shape}")
print(f"Num classes    : {len(np.unique(labels)) - 1}")  # excluding background

Download all datasets

from huggingface_hub import snapshot_download

local_path = snapshot_download(
    repo_id="Tanishq165/HSI_Datasets",
    repo_type="dataset"
)
print(f"Downloaded to: {local_path}")

---

📐 Data Format

---

🔬 Applications & Use Cases

• Hyperspectral Image Classification (HSI) — pixel-wise or patch-based
• Land Use / Land Cover (LULC) Mapping
• Benchmarking deep learning architectures — CNNs, Vision Transformers (ViT), Mamba/SSMs, Graph Neural Networks
• Transfer learning & domain adaptation across HSI scenes
• Multi-sensor data fusion (e.g., HSI + LiDAR for Trento/Augsburg)
• Planetary surface analysis — NiliFossae, Utopia and Holden(Mars CRISM data)
• Spectral unmixing and abundance estimation

---

📊 Dataset Specifications

Complete reference for all 24 datasets:

Note: H × W = Height × Width (spatial dimensions). Labeled samples exclude background pixels (class 0).

---

🗂️ Dataset File Reference

Each folder contains .mat files. The loader auto-detects data vs GT by filename suffix:

---

🏷️ Class Names Reference

Indian_Pines (16 classes)

1. Alfalfa | 2. Corn-notill | 3. Corn-mintill | 4. Corn | 5. Grass-pasture | 6. Grass-trees | 7. Grass-pasture-mowed | 8. Hay-windrowed | 9. Oats | 10. Soybean-notill | 11. Soybean-mintill | 12. Soybean-clean | 13. Wheat | 14. Woods | 15. Buildings-Grass-Trees-Drives | 16. Stone-Steel-Towers

Pavia University (9 classes)

1. Asphalt | 2. Meadows | 3. Gravel | 4. Trees | 5. Painted metal sheets | 6. Bare Soil | 7. Bitumen | 8. Self-Blocking Bricks | 9. Shadows

Pavia Center (9 classes)

1. Water | 2. Trees | 3. Asphalt | 4. Self-Blocking Bricks | 5. Bitumen | 6. Tiles | 7. Shadows | 8. Meadows | 9. Bare Soil

Salinas (16 classes)

1. Brocoli_green_weeds_1 | 2. Brocoli_green_weeds_2 | 3. Fallow | 4. Fallow_rough_plow | 5. Fallow_smooth | 6. Stubble | 7. Celery | 8. Grapes_untrained | 9. Soil_vinyard_develop | 10. Corn_senesced_green_weeds | 11. Lettuce_romaine_4wk | 12. Lettuce_romaine_5wk | 13. Lettuce_romaine_6wk | 14. Lettuce_romaine_7wk | 15. Vinyard_untrained | 16. Vinyard_vertical_trellis

KSC (13 classes)

1. Scrub | 2. Willow swamp | 3. CP hammock | 4. Slash pine | 5. Oak/Broadleaf | 6. Hardwood | 7. Swamp | 8. Graminoid marsh | 9. Spartina marsh | 10. Cattail marsh | 11. Salt marsh | 12. Mud flats | 13. Water

Botswana (14 classes)

1. Water | 2. Hippo grass | 3. Floodplain grasses 1 | 4. Floodplain grasses 2 | 5. Reeds | 6. Riparian | 7. Firescar | 8. Island interior | 9. Acacia woodlands | 10. Acacia shrublands | 11. Acacia grasslands | 12. Short mopane | 13. Mixed mopane | 14. Exposed soils

Houston13 (15 classes)

2013 IEEE GRSS Data Fusion Contest | 2.5 m/pixel | 380–1050 nm

1. Healthy Grass | 2. Stressed Grass | 3. Synthetic Grass | 4. Trees | 5. Soil | 6. Water | 7. Residential | 8. Commercial | 9. Road | 10. Highway | 11. Railway | 12. Parking Lot 1 | 13. Parking Lot 2 | 14. Tennis Court | 15. Running Track

Houston18 (20 classes)

2018 IEEE GRSS Data Fusion Contest | 1 m/pixel | 380–1050 nm | Area: UH campus + surroundings

1. Healthy Grass | 2. Stressed Grass | 3. Artificial Turf | 4. Evergreen Trees | 5. Deciduous Trees | 6. Bare Earth | 7. Water | 8. Residential Buildings | 9. Non-residential Buildings | 10. Roads | 11. Sidewalks | 12. Crosswalks | 13. Major Thoroughfares | 14. Highways | 15. Railways | 16. Paved Parking Lots | 17. Unpaved Parking Lots | 18. Cars | 19. Trains | 20. Stadium Seats

Trento (6 classes)

1. Apples | 2. Buildings | 3. Ground | 4. Woods | 5. Vineyard | 6. Roads

Muufl (11 classes)

1. Trees | 2. Mostly grass | 3. Mixed ground surface | 4. Dirt and sand | 5. Road | 6. Water | 7. Buildings shadow | 8. Buildings | 9. Sidewalk | 10. Yellow curb | 11. Cloth panels

NiliFossae (9 classes) — Mars

1. Fe-Olivine | 2. Epidote | 3. Chlorite | 4. Bassanite | 5. Illite/Muscovite | 6. Mg-Carbonate | 7. Plagioclase | 8. Prehnite | 9. Serpentine

Utopia (9 classes) — Mars

1. Analcime | 2. Bassanite | 3. High-Ca Pyroxene | 4. Illite/Muscovite | 5. Low-Ca Pyroxene | 6. Mg-Smectite | 7. Monohydrated sulfate | 8. Plagioclase | 9. Prehnite

Holden (6 classes) — Mars

1. Analcime | 2. Plagioclase | 3. Prehnite | 4. High-Ca Pyroxene | 5. Serpentine | 6. Margarite

WHU-Hi-HongHu (22 classes)

1. Red roof | 2. Road | 3. Bare soil | 4. Cotton | 5. Cotton firewood | 6. Rape | 7. Chinese cabbage | 8. Pakchoi | 9. Cabbage | 10. Tuber mustard | 11. Brassica parachinensis | 12. Brassica chinensis | 13. Small Brassica chinensis | 14. Lactuca sativa | 15. Celtuce | 16. Film covered lettuce | 17. Romaine lettuce | 18. Carrot | 19. White radish | 20. Garlic sprout | 21. Broad bean | 22. Tree

WHU-Hi-HanChuan (16 classes)

1. Strawberry | 2. Cowpea | 3. Soybean | 4. Sorghum | 5. Water spinach | 6. Watermelon | 7. Greens | 8. Trees | 9. Grass | 10. Red roof | 11. Gray roof | 12. Plastic | 13. Bare soil | 14. Road | 15. Bright object | 16. Water

WHU-Hi-LongKou (9 classes)

1. Corn | 2. Cotton | 3. Sesame | 4. Broad-leaf soybean | 5. Narrow-leaf soybean | 6. Rice | 7. Water | 8. Roads and houses | 9. Mixed weed

Dioni (12 classes)

1. Dense Urban Fabric | 2. Mineral Extraction Sites | 3. Non Irrigated Arable Land | 4. Fruit Trees | 5. Olive Groves | 6. Coniferous Forest | 7. Dense Sclerophyllous Vegetation | 8. Sparce Sclerophyllous Vegetation | 9. Sparsely Vegetated Areas | 10. Rocks and Sand | 11. Water | 12. Coastal Water

Loukia (14 classes)

1. Dense Urban Fabric | 2. Mineral Extraction Sites | 3. Non Irrigated Arable Land | 4. Fruit Trees | 5. Olive Groves | 6. Broad Leaved Forest | 7. Coniferous Forest | 8. Mixed Forest | 9. Dense Sclerophyllous Vegetation | 10. Sparce Sclerophyllous Vegetation | 11. Sparsely Vegetated Areas | 12. Rocks and Sand | 13. Water | 14. Coastal Water

Berlin (8 classes)

1. Forest | 2. Residential | 3. Industrial | 4. Low Plants | 5. Soil | 6. Allotment | 7. Commercial | 8. Water

Augsburg (7 classes)

1. Forest | 2. Residential Area | 3. Industrial Area | 4. Low Plants | 5. Allotment | 6. Commercial Area | 7. Water

Pingan (10 classes)

1. Seawater | 2. Road | 3. Trees | 4. Floating pier | 5. Brick houses | 6. Steel houses | 7. Ship | 8. Car | 9. Concrete building | 10. Grass

Qingyun (6 classes)

1. Trees | 2. Car | 3. Asphalt road | 4. Concrete building | 5. Water | 6. Grass

Tangdoaowan (18 classes)

1. Rubber track | 2. Asphalt | 3. Grassland | 4. Ligustrum vicaryi | 5. Bare soil | 6. Populus | 7. Flagging | 8. Boardwalk | 9. Bulrush | 10. Coniferous pine | 11. Buxus sinica | 12. Ulmus pumila L | 13. Sandy | 14. Roof shadows | 15. Gravel road | 16. Spiraea | 17. Photinia serrulata | 18. Seawater

Chikusei (19 classes)

1. Water | 2. Bare soil (farmland) | 3. Bare soil (park) | 4. Bare soil (roadside) | 5. Pavement (asphalt) | 6. Pavement (brick) | 7. Farmland (paddy) | 8. Farmland (other) | 9. Greenhouse | 10. Grass (park) | 11. Grass (roadside) | 12. Tree (park) | 13. Tree (roadside) | 14. Forest | 15. Building (low-rise) | 16. Building (high-rise) | 17. Building (factory) | 18. Power line | 19. Swimming pool

---

⚠️ Data Loading Notes

• Auto-transpose: The loader automatically detects and transposes data to (H, W, Bands) if needed.
• Background class: Class 0 is always background/unlabeled and excluded from training.
• v7.3 MAT files: Some datasets use HDF5-based .mat format. The loader falls back to h5py automatically.
• Class imbalance: Many datasets have highly imbalanced classes (e.g. Indian Pines, Houston18). Consider weighted loss or balanced sampling.

---

📄 Citation

If you use this dataset collection in your research, please cite:

@misc{rachamalla2026hsi,
      title={Hyperspectral Image Models: Technical Report}, 
      author={Tanishq Rachamalla and Aryan Das and Srishti Kaushik and Swalpa Kumar Roy},
      year={2026},
      eprint={2609.39871},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2609.39871}, 
}

Please also cite the original dataset papers for each specific dataset you use in your work.

---

🤝 Acknowledgements

These datasets are sourced from publicly available repositories and the broader remote sensing research community. Full credit goes to the original dataset creators including:

• NASA JPL / JPL AVIRIS — Indian Pines, Salinas, KSC, Botswana
• Wuhan University — WHU-Hi series (HanChuan, HongHu, LongKou)
• IEEE GRSS Data Fusion Contest — Houston 2013, Houston 2018
• University of Pavia — Pavia Center, Pavia University
• NASA MRO CRISM team — NiliFossae, Utopia, Holden (Mars datasets)
• DLR / HyMap — Berlin, Augsburg
• University of Southern Mississippi — Muufl

---

📬 Contact

For questions, issues, or collaboration, feel free to open a discussion on the HuggingFace dataset page.