OrbitalCADENCE

Content-Aware Data Compression for Multispectral Environments

Project Snapshot

Status

Finalist of the OrbitalAI Φ-sat-2 Challenge

Start Date

15/09/2023

End Date

2024

Programme

Content-Aware Data Compression for Multispectral Environments

Partners

OHB Hellas role

Prime

The Objective

Building on our success as a Finalist in the ESA OrbitalAI Challenge which involves:

  • Modernization: Transitioning to TensorFlow v2.x and TFLite.
  • Hardware Acceleration: Deploying and benchmarking on space-representative hardware including the Google Coral TPU and the Xilinx Kria KV260 (FPGA-based).
  • Performance Optimization: Improving execution time (targeting < 4 seconds per 256×256 tile) and optimizing the SCC (Spearman’s Rank Correlation) metrics.

OHB Hellas Contributions

OHB Hellas is the Prime Contractor for OrbitalCADENCE, a revolutionary content-aware algorithm that applies selective lossy compression. Our innovation lies in the “Intelligence-First” approach:

  1. Classification: An on-board CNN identifies Areas of Interest (AoI) such as land cover or vessel detection.
  2. Selective Compression: Areas of interest are compressed at low ratios (e.g., CR=4), while non-critical areas (clouds) are compressed at high ratios (e.g., CR=16).
  3. VQ-VAE Architecture: We utilize a Vector Quantized Variational AutoEncoder to transform high-dimensional data into a lower-dimensional latent space.

Related project in Orbital High-Performance Computing