CNN-based Star Tracker For High Precision Spacecraft Navigation
Date Issued
2024-04
Author(s)
Guliyev, Farid
Abstract
There are several sensors to detect the attitude determination of the spacecraft. One of
the most accurate is the star tracker which is optical equipment that measures the positions of
stars. However traditional star identification algorithms are limited by their processing speed
and may fail under harsh conditions which can be critical for the spacecraft. Furthermore a
large and complex database of star trackers can cause inefficiencies throughout the processing
process. An improved star tracker algorithm is proposed to address the issues above. This
algorithm will be able to handle these challenging situations and give the most accurate
estimation of the spacecraft orientation using state-of-the-art detection.
This paper proposes an improved star detection algorithm for spacecraft attitude
detection using convolutional neural networks (CNN), a type of deep learning algorithm that
has shown many promising results in various image processing applications. CNN-based star
detection algorithm will be able to identify the stars based on the constellations and show the
spacecraft's orientation using a less complex database. This will positively affect the accuracy
and efficiency of the processing and increase robustness against challenging conditions. Using
CNN-based algorithms creates an opportunity continuously improve spacecraft attitude control
systems.
Free, open-source databases are being used to train the algorithm to overcome the issue
of inaccessible star and constellation images from satellites. Publicly available data from
planetariums such as Stellarium and space simulator platforms such as Celestia are being used
for training, evaluation, and performance analysis of the proposed algorithm. Moreover various
optimization techniques are implemented to improve the performance of the spacecraft attitude
determination algorithm. It is believed that this approach will surpass the performance of the
traditional algorithm and will lead to a promising direction for the development of advanced
star tracker systems. Further research can focus on applying this approach to other areas of
spacecraft control such as orbit determination and maneuver planning.
the most accurate is the star tracker which is optical equipment that measures the positions of
stars. However traditional star identification algorithms are limited by their processing speed
and may fail under harsh conditions which can be critical for the spacecraft. Furthermore a
large and complex database of star trackers can cause inefficiencies throughout the processing
process. An improved star tracker algorithm is proposed to address the issues above. This
algorithm will be able to handle these challenging situations and give the most accurate
estimation of the spacecraft orientation using state-of-the-art detection.
This paper proposes an improved star detection algorithm for spacecraft attitude
detection using convolutional neural networks (CNN), a type of deep learning algorithm that
has shown many promising results in various image processing applications. CNN-based star
detection algorithm will be able to identify the stars based on the constellations and show the
spacecraft's orientation using a less complex database. This will positively affect the accuracy
and efficiency of the processing and increase robustness against challenging conditions. Using
CNN-based algorithms creates an opportunity continuously improve spacecraft attitude control
systems.
Free, open-source databases are being used to train the algorithm to overcome the issue
of inaccessible star and constellation images from satellites. Publicly available data from
planetariums such as Stellarium and space simulator platforms such as Celestia are being used
for training, evaluation, and performance analysis of the proposed algorithm. Moreover various
optimization techniques are implemented to improve the performance of the spacecraft attitude
determination algorithm. It is believed that this approach will surpass the performance of the
traditional algorithm and will lead to a promising direction for the development of advanced
star tracker systems. Further research can focus on applying this approach to other areas of
spacecraft control such as orbit determination and maneuver planning.