Imagine a future where your smartphone could help track meteors, satellites, or other objects in space and where supercomputers can process that data to keep our skies safer. This vision is becoming reality through a project led by the Luxembourg Institute of Science and Technology in cooperation with Royal Belgian Institute for Space Aeronomy.
Dubbed NEOD2 for AI for Space Objects Detection 2, the project combines citizen science, radio astronomy, and artificial intelligence to detect and monitor near-Earth objects. At the heart of the projects lies MeluXina, Luxembourg’s national supercomputer.
This project addresses a growing societal need as Earth’s orbit becomes increasingly crowded with satellites and debris, monitoring space objects is vital for space safety, protecting satellites, and avoiding collisions that could impact critical communications, weather forecasting, and navigation systems.
AI for near earth object detection
The original NEOD project focused on training generative AI models to automatically identify signals of meteors from radio data, tackling a long‑standing challenge in astronomy: separating meaningful signals from noise caused by atmospheric interference, sensor artefacts, and human activity. The models developed under NEOD are designed to be robust, scalable, and capable of running on distributed computing architectures.
Building on the achievements of it, NEOD2 expands both the scope and ambition of space object AI detection. It aims to apply and finetune AI models across a much larger dataset of radio echoes collected by the BRAMS network, in collaboration with Hervé Lamy and Stijn Calders (Royal Belgian Institute for Space Aeronomy), and thanks to the work carried out by Diogo Ramalho Fernandes and Mickael Stefas. One key objective is to generate an extensive database of detected meteor echoes, paired with trajectory and velocity reconstructions, to support operational readiness of the BRAMS network.
Olivier Parisot, Principal Investigator of the project at LIST, stated: “A novel component of NEOD2 involves integrating citizen science on the radio side as well. Data with lower confidence AI detections will be presented through Radio Meteor Zoo 2.0, a future version of a community curated platform where volunteers help validate or refine automated detections. This hybrid human–AI workflow not only improves detection accuracy but also deepens public engagement in frontline scientific research.”
The power behind the models: MeluXina
AI models of this complexity demand serious computational muscle. Enter MeluXina, Luxembourg’s national supercomputer operated by LuxProvide.
Olivier Parisot said: “Meluxina gives us valuable insight as it processes massive datasets, trains deep learning models, and conducts large‑scale benchmarking that would be impractical on standard hardware. The system plays a vital role in accelerating the transition from raw data to validated scientific insight.”
Scientific contributions and broader impacts
The projects have already generated several scientific publications that document their methodology and findings. Among them Toward a Public Dataset of Wide-Field Astronomical Images Captured with Smartphones, and Robustness Analysis of Deep Sky Objects Detection Models on HPC.
Beyond academic output, NEOD and NEOD2 address a growing global need for space situational awareness. As Earth’s orbit becomes more crowded with satellites and debris, the ability to detect and monitor objects is essential for collision avoidance, infrastructure protection, and long-term sustainability of space activities.
Through LIST’s distinctive model, results from fundamental research can be translated into applied tools and potential services for industry and public authorities. This capacity to bridge science and real-world application is what makes the institute strategically important for Luxembourg and its international partners.
Source: LIST (Luxembourg Institute of Science and Technology)