High-Accuracy Guide Star Catalogue Generation with a Machine Learning Classification Algorithm

Optimal uniform guide star catalog using a genetic algorithm

To achieve optimal and reliable star sensors and overcome some onboard hardware and software limitations, this study aimed to make an optimal uniform guide star catalog. For this purpose, the objective function was defined by the field of view (FOV) and magnitude threshold, and then design variables were optimized. The optimal uniform guide star catalog was obtained by a genetic algorithm alongside the Latinized stratified sampling method and by a novel, to the best of our knowledge, spherical density determination algorithm based on the minimum number of stars required for a star identification algorithm. Finally, Monte Carlo simulation was used to validate the results, which indicate a dramatic improvement, including a reduction in the number of stars in the uniform catalog and an increase in the probability of observing the minimum required stars for the star identification algorithm (at least 5 stars) in 98.34% of all possible optimal FOVs (about 12°).

Construction of the Guide Star Catalog for Double Fine Guidance Sensors Based on SSBK Clustering

In the Chinese Survey Space Telescope (CSST), the Fine Guidance Sensor (FGS) is required to provide high-precision attitude information of the space telescope. The fine star guide catalog is an essential part of the FGS. It is not only the basis for star identification and attitude determination but also the key to determining the absolute attitude of the space telescope. However, the capacity and uniformity of the fine guide star catalog will affect the performance of the FGS. To build a guide star catalog with uniform distribution of guide stars and catalog capacity that is as small as possible, and to effectively improve the speed of star identification and the accuracy of attitude determination, the spherical spiral binary K-means clustering algorithm (SSBK) is proposed. Based on the selection criteria, firstly, the spherical spiral reference point method is used for global uniform division, and then, the K-means clustering algorithm in machine learning is introduced to divide the stars into several disjoint subsets through the use of angular distance and dichotomy so that the guide stars are uniformly distributed. We assume that the field of view (FOV) is 0.2° × 0.2°, the magnitude range is 9∼15 mag, and the threshold for the number of stars (NOS) in the FOV is 9. The simulation shows that compared with the magnitude filtering method (MFM) and the spherical spiral reference point brightness optimization algorithm (SSRP), the guide star catalog based on the SSBK algorithm has the lowest standard deviation of the NOS in the FOV, and the probability of 5∼15 stars is the highest (over 99.4%), which can ensure a higher identification probability and attitude determination accuracy.