HYPSO-1: 179 Hyperspectral Satellite Images with Sea-Land-Cloud Labels

179 calibrated hyperspectral images from the HYPSO-1 satellite for Earth and ocean observation. The dataset includes pixel-level labels for sea, land, and cloud categories on 38 images, collected primarily between June and December 2022. Jon Alvarez published this reformatted version, which compresses raw data from about 130GB to 17GB.

Use Cases

• Training semantic segmentation models for sea, land, and cloud classification based on the provided pixel-level labels.
• Analyzing hyperspectral signatures for Earth and ocean observation applications mentioned in the description.
• Benchmarking remote sensing algorithms using the calibrated L1A format image files.
• Generating up-to-date radiance data for analysis using the referenced hypso Python package.

Strengths

• Includes 179 calibrated hyperspectral images, providing a substantial collection for analysis.
• Offers pixel-level semantic labels for sea, land, and cloud categories on 38 images.
• Data is reformatted and compressed, reducing total size from about 130GB to about 17GB for easier handling.
• Includes a trained segmentation model (.h5 file) from a related publication for reference or further development.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Label coverage is partial, with only 38 of the 179 images having pixel-level annotations.
• Freshness should be verified; the last metadata update was 2026-05-03, but primary image data is from 2022-2023.

Provenance

Source

HYPSO-1 satellite.

Collection Method

Satellite remote sensing.

Time Range

Primarily June to December 2022, with additional images from February and March 2023.

Freshness

Last updated 2026-05-03 10:10:21; primary image data from June-December 2022 with some from February-March 2023.