The Sea-Land-Cloud Ground-Truth Labeled Dataset

We introduce a diverse dataset of 200 calibrated Hyperspectral Images for Earth and ocean observation, including pixel-level labels for 38 images for sea/land/cloud categories. In the next table, we present the labeled data. The first column introduces the RGB composite of uncalibrated raw Hyperspectral Images, followed by the second column where we overlay the ground-truth labels onto the composite. We use blue for water systems (sea class), orange for land cover (land class), gray for cloudy/overexposed regions (clouds class), and red for the class boundaries. For every labeled scene, we offer additional metadata including continent, country, approximate geographical coordinates obtained from the on-board ADCS telemetry, scene type, date, solar zenith angle, exposure time, and the overexposed spectra ratio. Following the first table that shows the labeled subset for sea/land/clouds, we then present an overview of the rest of the dataset in the subsequent table. Every capture possesses an ID assigned for internal identification purposes, as well as for reference when addressing each specific capture. Click the blue buttons with a white arrow in the column header of the tables to sort the data at your convenience.

Although the raw, radiance and labels have been finalized and revised without any intended future changes, we continue to conduct additional verifications for the metadata associated with each capture. This includes ensuring, for instance, the precision of geographical location metadata for each capture. Please, note that the metadata is preliminary and hence susceptible to modifications throughout the months of September and October 2023. If you have any questions, feel free to contact us.

**: waiting to be filled.
__Current Status: we provide 180/200 images; 20 remaining scenes available in the next days.