Enhancing Star Tracker Accuracy with CNNs

This research introduces a CNN-based approach tailored for CubeSat star trackers, addressing the challenges posed by high sensor noise and stray light. The method employs a novel binary segmentation map and a distance map for improving centroid calculations, significantly outperforming traditional algorithms.

Key Features:

• Utilization of CNN for star pixel detection and centroid computation.
• Development of binary segmentation and distance maps for accurate centroiding.
• Capability of real-time execution on low-power Edge AI processors.

Significance:

The integration of CNNs into star tracking systems represents a pivotal shift towards more accurate celestial navigation instruments. This approach not only enhances the precision but also the robustness of star trackers, especially in adverse conditions, crucial for advancing CubeSat technologies.