TL;DR
Researchers at DARe, the defence AI research centre of the Alan Turing Institute, and the University of Birmingham have built an AI model that identifies individual parts of a satellite in high-resolution radar images taken in orbit. The work was presented at the 2026 European Radar Conference on Friday 9 October. The aim is automated characterisation of objects in space, a field known as space domain awareness.
What the model does
Given radar imagery of a spacecraft, the model separates out and labels parts like solar panels, antennas, thrusters and robotic arms. It also works through images in sequence, a technique the Turing calls temporal memory encoding. By carrying forward what it saw in earlier frames, it can keep track of a part even after the satellite turns and the part is hidden from view, which matters because objects in orbit are always moving.
The imagery comes from a novel sub-terahertz inverse synthetic aperture radar, developed by Professor Marina Gashinova’s Pervasive Sensing Group at Birmingham. Its resolution is what makes component-level identification possible. The Turing says the work could help with safe docking and rendezvous, which are needed to service satellites and remove debris.
Why it matters for the UK
Satellites support services the UK relies on, such as weather forecasting and GPS, and a more crowded orbit raises the risk of collisions. Monitoring thousands of active satellites by hand does not scale.
Dr Victoria Nockles, who heads DARe, says space domain awareness has traditionally depended on ground-based radar and telescopes, and that demand is growing to watch debris and satellites from orbit itself. She describes the work as building “important sovereign capability”.
How it builds on earlier work
Nockles’ centre has tackled satellites before. Earlier, DARe published a tool, built with the University of Strathclyde, that learned from light curves (how light reflects off a satellite) collected by ground telescopes, and identified unusual readings 88% of the time. We covered it on 2 September as Turing AI spots satellite trouble from a single point of light.
Looking forward
The two projects tackle different halves of the same problem: spotting when something in orbit is behaving oddly, and working out what that object is made of. In our view, the radar work is the bigger step for sovereignty, because it relies on sensors in space rather than on ground observatories. The Turing has not said when the system might be used operationally, so this remains research rather than a deployed capability.