Raw-to-Processed in Orbit: Fast Onboard Image Reconstruction for EO

Gabriele Inzerillo, Lorenzo Papa, Gabriele Meoni, Diego Valsesia, Enrico Magli

Keywords: Onboard AI, Image Reconstruction, Earth Observation, Computer Vision

Abstract:

Onboard Artificial Intelligence (AI) is increasingly adopted in Earth Observation (EO) missions to reduce downlink requirements and enable low-latency decision-making.
However, converting raw acquisitions into usable products still relies on computationally intensive preprocessing steps that are typically executed on onboard CPU, creating a major bottleneck for resource-constrained CubeSats such as $\Phi$Sat-2.
In this work, we propose an end-to-end Deep Learning (DL) pipeline that reconstructs processed image products directly from raw data, explicitly designed for deployment on space-qualified AI accelerators.
Our approach combines (i) a lightweight learned multispectral band coregistration module and (ii) an efficient wavelet-domain image reconstruction network.
Experiments show that the proposed method achieves up to a $6.63\times$ latency reduction compared to the traditional onboard CPU-based preprocessing chain, while preserving reconstruction fidelity.
Overall, these results suggest that learned raw-to-processed reconstruction is a viable and efficient alternative to conventional onboard preprocessing chain, and provide a step toward fully AI-driven onboard EO pipelines.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 13