🌱 PASTIS-HD 🌿 Panoptic Agricultural Satellite TIme Series : optical time series, radar time series and very high resolution image

PASTIS is a benchmark dataset for panoptic and semantic segmentation of agricultural parcels from satellite time series. It contains 2,433 patches within the French metropolitan territory with panoptic annotations (instance index + semantic label for each pixel). Each patch is a Sentinel-2 multispectral image time series of variable lentgh.

This dataset have been extended in 2021 with aligned radar Sentinel-1 observations for all 2433 patches.
For each patch, it constains approximately 70 observations of Sentinel-1 in ascending orbit, and 70 observations in descending orbit. Each each Sentinel1 observation is assembled into a 3-channel image: vertical polarization (VV), horizontal polarisation (VH), and the ratio vertical over horizontal polarization (VV/VH). This extension is named PASTIS-R.

We extend PASTIS with aligned very high resolution satellite images from SPOT 6-7 constellation for all 2433 patches in addition to the Sentinel-1 and 2 time series. The image are resampled to a 1m resolution and converted to 8 bits. This enhancement significantly improves the dataset's spatial content, providing more granular information for agricultural parcel segmentation. PASTIS-HD can be used to evaluate multi-modal fusion methods (with optical time series, radar time series and VHR images) for parcel-based classification, semantic segmentation, and panoptic segmentation.

Dataset in numbers

🛰️ Sentinel 2	🛰️ Sentinel 1	🛰️ SPOT 6-7 VHR	🗻 Annotations
➡️ 2,433 time series	➡️ 2 time 2,433 time series	➡️ 2,433 images	124,422 individual parcels
➡️ 10m / pixel	➡️ 10m / pixel	➡️ 1.5m / pixel	covers ~4,000 km²
➡️ 128x128 pixels / images	➡️ 128x128 pixels / images	➡️ 1280x1280 pixels / images	over 2B pixels
➡️ 38-61 acquisitions / series	➡️ ~ 70 acquisitions / series	➡️ One observation	18 crop types
➡️ 10 spectral bands	➡️ 2 spectral bands	➡️ 3 spectral bands

⚠️ The SPOT data are natively 1.5m resolution, but we over-sampled them at 1m to align them pixel-perfect with Sentinel data.

Data loading

The Github repository associated to this dataset contains a PyTorch dataset class of the OmniSat repository that can be readily used to load data for training models on PASTIS-HD. The time series contained in PASTIS have variable lengths. The Sentinel 1 and 2 time series are stored in numpy array. The SPOT images are in TIFF format. The annotations are in numpy array too.

⚠️ The S2 and S1 folders contains more than 2433 files on the contrary to the labels folder. Some patches are not labelled and not used for training. The relevant information can be find in the metadata.geojson file (with 2433 entries), which is used as an index by the dataloader.

Remark about the folder names

⚠️ The DATA_S1A folder contains the Sentinel-1 ascendent images whereas the DATA_S1D folder contains the Sentinel-1 descendant images.

Ground Truth Annotations

The agricultural parcels are grouped into 18 different crop classes as shown in the table below. The backgroud class corresponds to non-agricultural land, and the void label for parcels that are mostly outside their patch.

Additional information about the dataset can be found in the documentation/pastis-documentation.pdf document.

Credits

• The Sentinel imagery used in PASTIS was retrieved from THEIA: "Value-added data processed by the CNES for the Theia www.theia.land.fr data cluster using Copernicus data. The treatments use algorithms developed by Theia’s Scientific Expertise Centres. "

• The annotations used in PASTIS stem from the French land parcel identification system produced by IGN.

• The SPOT images are opendata thanks to the Dataterra Dinamis initiative in the case of the "Couverture France DINAMIS" program.

References

If you use PASTIS please cite the related paper:

@article{garnot2021panoptic,
  title={Panoptic Segmentation of Satellite Image Time Series
with Convolutional Temporal Attention Networks},
  author={Sainte Fare Garnot, Vivien  and Landrieu, Loic},
  journal={ICCV},
  year={2021}
}

For the PASTIS-R optical-radar fusion dataset, please also cite this paper:

@article{garnot2021mmfusion,
  title    = {Multi-modal temporal attention models for crop mapping from satellite time series},
  journal  = {ISPRS Journal of Photogrammetry and Remote Sensing},
  year     = {2022},
  doi      = {https://doi.org/10.1016/j.isprsjprs.2022.03.012},
  author   = {Vivien {Sainte Fare Garnot} and Loic Landrieu and Nesrine Chehata},
}

For the PASTIS-HD with the 3 modalities optical-radar time series plus VHR images dataset, please also cite this paper:

@article{astruc2024omnisat,
  title={Omni{S}at: {S}elf-Supervised Modality Fusion for {E}arth Observation},
  author={Astruc, Guillaume and Gonthier, Nicolas and Mallet, Clement and Landrieu, Loic},
  journal={ECCV},
  year={2024}
}