Centroid algorithm for high-dynamic star sensor based on key point detection deep learning algorithms.

Area of Science:

• Spacecraft Attitude Determination
• Computer Vision
• Deep Learning

Background:

• Dynamic conditions cause star map trailing, reducing star sensor accuracy.
• Traditional methods struggle with blurred star maps and complex trailing patterns.
• Accurate star centroid positioning is crucial for satellite attitude determination.

Purpose of the Study:

• To develop a novel deep learning model for accurate star centroid positioning under dynamic conditions.
• To overcome limitations of traditional methods in handling star map trailing.
• To improve the robustness and accuracy of star sensor performance in complex scenarios.

Main Methods:

• A deep learning model based on object detection was employed.
• Star centroid positioning was reframed as key point positioning in trailing star maps.
• The model was compared against traditional centroid algorithms.

Main Results:

• Centroid positioning accuracy improved by over an order of magnitude.
• Successfully addressed positioning of non-uniform/non-linear trailing tracks and overlapping streaks.
• Demonstrated high robustness to noise in star images.

Conclusions:

• Introduced a new paradigm for centroid positioning in dynamically blurred star images.
• Pushed the upper limit of centroid positioning accuracy in high-dynamic scenarios.
• Provided a solution for precise attitude determination in complex satellite operational environments.