AI model picks out satellite parts from radar images
Researchers at the Alan Turing Institute and the University of Birmingham developed an artificial-intelligence model that identifies satellite components from high-resolution radar imagery. The model segments features including solar panels, thrusters, antennas and robotic arms, using temporal memory to retain information across image sequences as objects rotate or components become obscured. Its input imagery uses a sub-terahertz inverse synthetic aperture radar system developed at Birmingham, supporting research into autonomous characterization of orbiting objects. Presented at the European Radar Conference 2026 on October 9, the work could support space-domain awareness and future satellite docking, servicing and debris-removal operations.