Researchers have built an artificial intelligence model that can identify individual components on a satellite from high-resolution radar imagery taken in orbit, including solar panels, thrusters, antennas and robotic arms, according to the Alan Turing Institute.
The work comes from the Turing’s Defence AI Research Centre, led by Dr Victoria Nockles, alongside the University of Birmingham, and was presented on Friday at the European Radar Conference 2026.
Thousands of active satellites and other objects in orbit have to be monitored manually, and as the environment grows more congested the risk of collision rises with it. Satellites underpin services Britain depends on, from GPS to weather forecasting.
Temporal memory encoding allows the model to analyse radar images in sequence and learn from earlier ones to interpret later ones, so it can remember a component seen before and keep tracking it even when the satellite rotates and hides the part from view. Everything in orbit is in constant motion, and that persistence has been difficult to achieve.
The imagery comes from a high frequency inverse synthetic aperture radar system operating in the sub-terahertz band, developed by the Pervasive Sensing Group under Professor Marina Gashinova at Birmingham. Its resolution allows individual components to be picked out, and the same capability could support satellite docking and rendezvous, which matter for maintenance and debris removal.
Dr Nockles, principal data scientist and head of the Defence AI Research Centre, said: “Traditional space domain awareness relies on the ground-based infrastructure of both radar and telescopes. However, there is increasing demand to monitor and characterise satellites and debris from space. Our work is building important sovereign capability to safeguard space assets using novel radar imaging technology onboard satellites to observe space objects in ways that aren’t possible from the ground. This research is at the frontier of three strategic areas: space domain awareness, AI, and autonomy.”
one can imagine how Russia which, currently has to rely on salvaged washing machine parts to build missiles, will be in a position to fight NATO in a full-scale AI led war by 2030, when we are doing stuff like this and they are relying on meat wave assaults on e scooters.