CNN-Based Approach for Star Tracking

Discover the advancements in celestial navigation using a real-time convolutional neural network (CNN) designed for CubeSat star trackers. This paper introduces a method that significantly enhances the detection accuracy of the system under challenging conditions such as high sensor noise and stray light. Here’s what it entails:

• Use of CNN for binary segmentation and distance map generation.
• Implementation of trilateration based on pixel coordinates for accurate centroid calculations.
• UNet variants evaluation and comprehensive tests under various conditions.

Why this is significant: The technology allows for real-time processing on low-power edge AI processors and offers a substantial improvement over traditional methods in terms of accuracy and robustness. The integration of advanced machine learning techniques in space technology opens new avenues for research in optimally configuring sensors and algorithms for space missions.