Preprint available!
I am proud to announce the availability of our new article “Explaining raw data complexity to improve satellite onboard processing”.
https://arxiv.org/abs/2510.06858
In this paper, we aim to reduce the gap between satellite data and intelligent behaviour by examining how raw sensor data affects the performance of AImodels.
To achieve this, we trained lightweight detection models specifically designed for deployment on an FPGA with the task of detecting vessels in high-resolution raw imagery. The models were trained on images simulated to mimic the characteristics of a sensor’s output. Explainability XAI is then leveraged to dive deeper into the challenges faced by AI models.
The results suggest that the computational cost of onboard preprocessing can be significantly reduced, marking a game-changerin accelerating onboard AI processing for space missions.
For more information, I will be presenting the paper at the EDHPC conference in Spain!
A big thanks to Centre National d’Études Spatiales and IRT Saint Exupéry for their support in this work.