AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-23
SIGNN++: Robust Graph Neural Star Identification on Earth
Learning-based star identification for night-time celestial navigation without GNSS had mostly been tested on synthetic or non-terrestrial data. The authors extended the graph neural network method SIGNN with a more realistic noise model, Fourier transform of angular-distance edge features, and RANSAC-based perspective star alignment. The result was greater robustness to terrestrial observation disturbances, with higher identification accuracy and solved rates than SIGNN and public baselines, while attitude accuracy and computation time stayed comparable.