onboard-coop/STARCOP-fast-products

Paper | GitHub Repository

Overview

This repository hosts curated data products used for evaluation and training in the article "Optimizing Methane Detection Onboard Satellites: Speed, Accuracy, and Low-Power Solutions for Resource-Constrained Hardware."

The dataset consists exclusively of products generated from 72 hyperspectral channels within the wavelength range of 2122–2488 nm. These products served as the training data for the models available here.

The original raw hyperspectral data required to create these products is publicly accessible at: https://huggingface.co/collections/previtus/starcop-67f13cf30def71591f281a41

Sample Usage (Notebook Demos)

You can try out our demos directly in Google Colab:

• —<a href="https://colab.research.google.com/github/zaitra/methane-filters-benchmark/blob/main/ntbs/Models_demo.ipynb"> Models Demo <img src="https://colab.research.google.com/assets/colab-badge.svg" height=16px></a> Demonstrates model inference.

• —<a href="https://colab.research.google.com/github/zaitra/methane-filters-benchmark/blob/main/ntbs/Products_demo.ipynb"> Products Creation and Benchmarking Demo <img src="https://colab.research.google.com/assets/colab-badge.svg" height=16px></a> Demonstrates generating products and measuring their runtime.

Citation

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

@misc{herec2026fastmethanedetectionpipeline,
      title={A Fast Methane Detection Pipeline on Board Satellites Based on Mag1c-SAS and LinkNet}, 
      author={Jonáš Herec and Vít Růžička and Rado Pitoňák and Jan Sedmidubsky},
      year={2026},
      eprint={2606.03675},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2606.03675}, 
}