CORSA

A family of lightweight AI Compression and Foundation Models for Multi- and Hyperspectral Data.

Revolutionizing EO with AI-driven Data Compression

Developed by VITO Remote Sensing, CORSA revolutionizes data compression and processing for Earth Observation.
Using advanced deep learning, it efficiently compresses imagery from sensors like Sentinel-1, Sentinel-2 and PRISMA, drastically reducing data volume while maintaining high image fidelity, even at 100x compression rates.

CORSA's flexibility offers tailored solutions, balancing compression ratio and reconstruction quality to meet specific project needs. It provides custom trained models for new sensors, pretrained models for optimizing EO portal storage, and AI training templates for downstream applications.
CORSA's compressed features can be directly used to build downstream applications, enhancing the development and accessibility of hyperspectral application despite data transmission limitations.

This technology enhances efficiency and reduces costs for space providers and space agencies.
CORSA paves the way for advanced AI in future space missions and will expand into hyperspectral data analyses, broadening its impact across EO sectors.

Key Features

AI-Driven Compression: implements a novel deep learning architecture to compress optical EO imagery into compact vectors.
Edge Computing: reduces data transmission needs by processing data closer to the source.
Self-Supervised Learning: models learn image representations without needing human-labeled data, adaptable on-board with incremental finetuning.

Benefits

Efficient Data Use: greatly reduces the volume of data that needs to be transmitted and stored, up to 100x compression.
High-Quality Reconstruction: maintains high image fidelity with minimal quality loss.
Versatile Application: supports a range of downstream tasks like land use classification change detection and natural disaster mapping directly from compressed data.

Offered services

Custom Trained Models for New Sensors: tailored for efficient onboard integration with a custom ground decoder.
Pretrained Models for EO Portal: optimizes large-scale EO data storage and processing.
Downstream Application AI: provides templates for training algorithms on compressed vectors.
NEW!
Prototype to compress Sentinel-2 images with the CORSA algorithm
- via the notebook at the CORSA Github
- via Terrascope