Robust star identification using topology-invariant features and a SOM neural network

Area of Science:

• * Spacecraft engineering and astrodynamics.
• * Artificial intelligence and machine learning applications in aerospace.

Background:

• * Software-defined satellites (SDS) offer enhanced flexibility and reusability for imaging payloads through in-orbit reconfiguration.
• * Traditional star identification methods struggle with SDS due to fixed camera parameters, limiting robustness in dynamic environments.
• * A universal, adaptive algorithm is crucial for stable attitude determination in SDS architectures.

Purpose of the Study:

• * To develop a parameter-independent star identification algorithm for SDS.
• * To ensure robust and accurate attitude determination across diverse operational conditions.
• * To support the flexible and reconfigurable nature of SDS missions.

Main Methods:

• * Proposed a star identification algorithm based on topological invariance features.
• * Integrated a topological invariant feature coordinate system with a self-organizing map (SOM) neural network.
• * Tested the algorithm under various noise conditions and dynamic parameters (stellar position noise, false-star noise, missing-star noise, focal length variation, dynamic fields of view).

Main Results:

• * The proposed algorithm consistently outperformed three benchmark methods in extensive experiments.
• * Demonstrated stable accuracy across diverse payload configurations and observation scenarios.
• * Validated the algorithm's robustness against various noise types and parameter variations.

Conclusions:

• * The developed algorithm aligns with the SDS philosophy of hardware-software decoupling.
• * Provides a foundational algorithm for flexible, reconfigurable, and multi-payload attitude determination in SDS.
• * Enhances the reliability and adaptability of star identification for next-generation satellites.