General Method of the Automatic Generation of Onboard Triplet

Abstract

A novel general method of the automatic selection of onboard star triplet, namely triplet regression selection algorithm (TRSA), which based on a new dynamical label visual magnitude threshold (DLVMT) model, is presented. By defining the label visual magnitude and the direction of the star triplet, the star triplet distribution is analyzed. Using the DLVMT to filter the star triplet set, a new catalog with uniform distribution of the triplets over the celestial sphere can be obtained. The DLVMT distribution function has been attained via the support vector machines (SVM) regression method. With the proposed sampling method, computer experiments were carried out. The experiment results demonstrate that the triplet database obtained by the proposed algorithm has a couple of advantages, including fewer total numbers, smaller catalog size, and better distribution uniformity. 1 with which their operation capability to cover most or even all mission phases can be widened and all attitude data required for control can be supplied. The development of full autonomy of operation is also in accordance with the requirements of saving power, mass, and volume, and of limiting complexity and redundancy of onboard systems. An autonomous star tracker can operate and manage independently different mission phase requirements without support from other spacecraft units except the star image. These phases include the start up routine to determine the rough localization of the observed region of the sky, and the normal tracking mode following the initial acquisition procedure to estimate the high-precision attitude of the spacecraft. These different specific features are usually attained via software procedures. To obtain full autonomous attitude estimation, the star tracker should perform a prompt identification of the viewed star field by comparing observed star features and star characteristics stored in its onboard catalog. Once a correct match is made, there are reliable methods for generating good attitude estimation. Recently, many star pattern recognition (SPR) algorithms to generate a best match between the measured star pattern in the FOV and the subimage of the onboard catalog have been proposed. According to their respective identification approaches in the FOV, these algorithms can be divided into three classes. The first class of algorithms is the inter-star pair that has angular separation-based matching methods, in which the stars are treated as vertexes in a graph whose edges correspond to the angular separation between neighboring stars that could possibly share the same sensor FOV, such as those from Refs. 1 and 2. The grid algorithms, such as those from Refs. 3 and 4 belong to the second class of algorithms, in which the well-defined pattern determined by the surrounding star field has been associated with every star. The third class of algorithms is the developing neural networks-based recognition algorithms, 5 in which the star images of the FOV are treated as patterns that can be recognized directly. Because the neural network structure itself contains the information about the star feature vectors, the precompiled star feature database is not necessary to the neural networks-based star identification strategies. Except