Efficient Star Identification Using a Neural Network

Area of Science:

• Spacecraft attitude determination
• Sensor technology
• Artificial intelligence in aerospace

Background:

• Increasing precision requirements for spacecraft attitude determination necessitate advanced sensors.
• Star trackers offer arc-second precision and are becoming smaller, faster, and more efficient for micro-satellites.
• Lost-in-space star identification algorithms are critical for autonomous attitude determination without prior information.

Purpose of the Study:

• To present an efficient lost-in-space star identification algorithm for spacecraft.
• To leverage neural networks and a novel feature extraction method for improved performance.
• To achieve constant O(1) search time for star identification.

Main Methods:

• Development of a novel feature extraction method.
• Implementation of a neural network for implicit pattern storage.
• Elimination of database lookup in the star matching process.

Main Results:

• Achieved constant O(1) search time, independent of stored patterns.
• Demonstrated unrivalled search speed compared to other star identification algorithms.
• The algorithm exhibits excellent performance in a simple and lightweight design.

Conclusions:

• Neural networks are a preferred choice for star identification algorithms due to their efficiency and performance.
• The proposed algorithm meets the growing demand for accurate and fast attitude determination sensors.
• The lightweight design is suitable for micro-satellite applications.