HYPSO-1 Sea-Land-Cloud-Labeled Dataset

The HYPSO-1 Sea-Land-Cloud-Labeled Dataset is an open-source dataset that contains 200 diverse hyperspectral images captured by the HYPSO-1 satellite mission. This dataset can be a valuable resource for researchers and students working in Earth observation or interested in applying machine learning to satellite imagery.

You can find the dataset, related software and documentation by following this link.

Traditionally, the lack of labeled datasets has been a major hurdle in using satellite hyperspectral imaging (HSI) data for AI applications. However, with our contribution, we've solved this problem. We've labeled 38 of the images, covering a range of global locations, and categorized each element as sea, land, or cloud, resulting in a dataset of over 25 million labeled spectral signatures.

The figure shows RGB-renders of hyperspectral captures and their respective classes.

To showcase the potential of the dataset, we've optimized a deep learning model specifically designed for this task. The 1D Fully Convolutional Network achieves state-of-the-art performance in sea-land-cloud segmentation from hyperspectral imagery.

The entire dataset, including labeled data, the deep learning model, and all the software code used, is freely available for download. We've also provided a supplementary website where you can access everything you need to get started. You can find the dataset's accompanying research paper titled "An Open Hyperspectral Dataset with Sea-Land-Cloud Ground-Truth from the HYPSO-1 Satellite" on arXiv, along with the manuscript itself (also in PDF) through the provided link.

We believe this Hyperspectral Dataset will be a valuable asset for the HSI community and researchers in Earth and ocean observation. If you have any questions or feedback, please do not hesitate to contact us.