RPNet: A Representation Learning-Based Star Identification Algorithm

Likai Xu,Lei Liu,Jie Jiang

doi:10.1109/access.2019.2927684

Abstract

A novel star identification network (RPNet) based on representation learning is proposed in this paper. Unlike other pattern-based stars identification algorithms, the RPNet does not require the creation of an elaborate pattern, nor does it need to search among patterns. Instead, a star pattern generator (SPG) in the RPNet helps in finding the best pattern that can distinguish different stars clearly. A star pattern classifier (SPC) in the RPNet is utilized to recognize the pattern generated before. The simulations show that the RPNet is extremely robust toward star position noise, star magnitude noise, and false stars. The performance on simulation images outperforms almost all other pattern-based stars identification algorithms. On average, it achieves an identification rate of 99.23% in simulated star images. The identification rate on real star images is higher than 98%. Moreover, the algorithm achieves this performance with lesser memory and faster speed compared to polygon algorithms.

Highlights

The navigation system is an indispensable part of a spacecraft and includes attitude determination, speed measurement, and positioning
Though most of the time it works on tracking mode, a star identification process is needed for a star sensor to obtain current attitude of the spacecraft once it loses in space
A weight searching strategy is proposed in the paper for filtering and verification of guiding reference star (GRS) identified by representation learning based neural network (RPNet), which improves tremendously the identification ability for a single star image

Summary

Introduction

The navigation system is an indispensable part of a spacecraft and includes attitude determination, speed measurement, and positioning. The star sensor [1] is an important device in the navigation system that can determine its three-axis attitude without any prior information. The star sensor has two operating modes: The initial attitude acquiring mode and tracking mode. Though most of the time it works on tracking mode, a star identification process is needed for a star sensor to obtain current attitude of the spacecraft once it loses in space. Star identification algorithms aim at finding the correlation between observed and cataloged stars. It mainly falls into two categories: the subgraph isomorphism algorithm and patternbased identification algorithm

Methods

Results

Conclusion