Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Robust star identification using a neural network
Thesis (MEng)--Stellenbosch University, 2023.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | en_ZA en_ZA |
| Published: |
Stellenbosch : Stellenbosch University
2023
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613803734106112 |
|---|---|
| access_status_str | Open Access |
| author | Burger, Andries |
| author2 | Jordaan, Willem |
| author_browse | Burger, Andries Jordaan, Willem |
| author_facet | Jordaan, Willem Burger, Andries |
| author_sort | Burger, Andries |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2023. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/127276 |
| institution | Stellenbosch University (South Africa) |
| language | en_ZA en_ZA |
| last_indexed | 2026-06-10T12:41:57.021Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2023 |
| publishDateRange | 2023 |
| publishDateSort | 2023 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/127276 Robust star identification using a neural network Burger, Andries Jordaan, Willem Visagie, Lourens Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Artificial satellites -- Attitude control systems Star trackers Neural networks (Computer science) Stellar dynamics Thesis (MEng)--Stellenbosch University, 2023. ENGLISH ABSTRACT: Star trackers are used to determine a satellite’s attitude in space by capturing an image of the stars that can be seen from the satellite. This attitude is vital for the control of the satellite. The star tracker’s measured attitude is used by the attitude control system for fine pointing and control of the satellite. However, conventional identification techniques require accurate calibration, and suffer from decreased accuracy or extended execution times in the presence of false stars and noisy images. This project aims to address the shortcomings and limitations of conventional star identification techniques by applying a neural network to the star identification problem. A neural network was chosen because of the improvements that have been made to image classification problem with the application of neural networks. The conventional identification techniques use a feature that is extracted from the image and compared to a catalogue of matching features. To understand the effectiveness of the matching features, data set analysis was performed on the feature catalogue. This analysis was performed on two common matching features: Angular Distance (AD) and star discretisation. This analysis showed that both matching features perform best with a higher concentration of stars. However, when AD is used, the tracker’s identification accuracy decreases as the size of the feature catalogue increases. This analysis also showed that AD is sensitive to noise, and that discretisation is sensitive to both false and undetected stars. Star images were simulated to train the neural network. These simulated images were used, along with several real star tracker captured images, to test and compare the proposed neural network to the implemented conventional identification techniques. The simulated image results showed that the proposed network improves upon the identification accuracy of the conventional techniques. The proposed network has a lower sensitivity to noise but is slightly more sensitive to false and undetected stars. The neural network performance using real star tracker images did not show the same improvement. This can be attributed to brightness variation between simulated images in the training set compared to real-world images. The noise applied to the training set did not sufficiently account for the real-world variation in brightness. Future work will involve better matching of the simulated training sets to star tracker images. Additionally, the results showed that a validation algorithm would further improve the network’s identification accuracy and reduce its sensitivity to noise, false and undetected stars. AFRIKAANS OPSOMMING: Sterkameras word gebruik om ’n satelliet se ori¨entasie in die ruimte te bepaal deur ’n beeld van die sterre wat vanaf die satelliet sigbaar is, te neem. Hierdie ori¨entasie is noodsaaklik vir die beheer van die satelliet. Die sterkamera se gemete ori¨entasie word deur die ori¨entasiebeheerstelsel gebruik vir fyn rigting wys en beheer van die satelliet. Konvensionele identifikasietegnieke vereis egter akkurate kalibrasie, en ly aan verminderde akkuraatheid of verlengde uitvoeringstye in die teenwoordigheid van vals sterre en ruiserige beelde. Hierdie projek het dit ten doel om die tekortkominge en beperkings van konvensionele ster-identifikasietegnieke aan te spreek deur die toepassing van ’n neurale netwerk op die ster identifikasie probleem. ’n Neurale netwerk is gekies as gevolg van die verbeterings wat in beeld klassifikasie probleme aangebring is met die toepassing van neurale netwerke. Die konvensionele identifikasie tegnieke gebruik ’n kenmerk wat uit die beeld onttrek is en vergelyk dit met ’n katalogus van bypassende kenmerke. Om die doeltreffendheid van die passing van kenmerke te verstaan, is datastel-ontleding op die kenmerkkatalogus uitgevoer. Hierdie ontleding is uitgevoer op twee algemene ooreenstemmende kenmerke: Hoekafstand (HA) en sterdiskretisering. Hierdie ontleding het aangedui dat beide ooreenstemmende kenmerke die beste presteer met ’n ho¨er konsentrasie sterre. Wanneer HA egter gebruik word, verminder die sterkamera se identifikasie akkuraatheid soos die kenmerkkatalogus vergroot. Hierdie ontleding het ook aangedui dat HA sensitief is vir ruis, en dat diskretisering sensitief is vir beide vals en onopgemerkte sterre. Sterbeelde is gesimuleer om die neurale netwerk op te lei. Hierdie gesimuleerde beelde is gebruik, saam met verskeie egte sterkamera afgeneemde beelde, om die voorgestelde neurale netwerk te toets en te vergelyk met die ge¨ımplementeerde konvensionele identifikasie tegnieke. Die gesimuleerde beeldresultate het aangedui dat die voorgestelde netwerk verbeter op die identifikasie-akkuraatheid van die konvensionele tegnieke. Die voorgestelde netwerk het ’n laer sensitiwiteit vir ruis, maar is effens meer sensitief vir vals-en onopgemerkte sterre. Die neurale netwerkprestasie, met die gebruik van egte sterkamera-beelde, het nie dieselfde verbetering aangedui nie. Dit kan toegeskryf word aan helderheid variasie tussen gesimuleerde beelde in die oefenstel in vergelyking met werklike beelde, waar die toegepaste ruis op die oefenstel nie voldoende vergelyk met die werklike variasie nie. Toekomstige werk sal beter ooreenstemming tussen die gesimuleerde oefenstelle met die sterkamerabeelde behels. Daarbenewens dui die resultate aan dat ’n valideringsalgoritme die netwerk se identifikasie akkuraatheid verder sal verbeter en die sensitiwiteit teenoor ruis, vals en onopgemerkte sterre sal verminder. Masters 2023-03-02T08:38:10Z 2023-05-18T07:13:34Z 2023-03-02T08:38:10Z 2023-05-18T07:13:34Z 2023-03-01 Thesis http://hdl.handle.net/10019.1/127276 en_ZA en_ZA Stellenbosch University xvii, 125 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Artificial satellites -- Attitude control systems Star trackers Neural networks (Computer science) Stellar dynamics Burger, Andries Robust star identification using a neural network |
| title | Robust star identification using a neural network |
| title_full | Robust star identification using a neural network |
| title_fullStr | Robust star identification using a neural network |
| title_full_unstemmed | Robust star identification using a neural network |
| title_short | Robust star identification using a neural network |
| title_sort | robust star identification using a neural network |
| topic | Artificial satellites -- Attitude control systems Star trackers Neural networks (Computer science) Stellar dynamics |
| url | http://hdl.handle.net/10019.1/127276 |
| work_keys_str_mv | AT burgerandries robuststaridentificationusinganeuralnetwork |