SIGNN++: A network of neurons on graph recognizes stars from the ground to navigate at night without GPS

Thought for celestial navigation when GNSS is unavailable, the method is more resistant than its predecessors to the disturbances of a terrestrial observation, at a comparable time of calculation.

When GNSS signals are scrambled or absent, stars remain a landmark: it is still necessary to reliably identify those seen by the camera. Learning-based identification methods have been evaluated mainly on synthetic data or in space, where the sky is clear; their reliability under the atmosphere, with its perturbations, remained poorly understood. Jeongbin Sohn, Jina Lee, Max Yen, Dowan Gwon and their colleagues present SIGNN++, an evolution of SIGNN, a graph-based neural network stellar identification method.

Three additions for the field

SIGNN++ provides a more realistic noise model to generate the training data, a Fourier transform applied to the angular distances that describe the edges of the graph, and a perspective star alignment by RANSAC to confirm the identification and refine the attitude estimation. The result: better robustness to Earth observation perturbations, a higher identification and resolution rate than SIGNN and public reference methods, comparable attitude accuracy, and computation time compatible with continuous operation. The code and datasets must be published.