Anti-noise Star Image Extraction Algorithm for Star Trackers Based on YOLOv5

The spacecraft depend on accurate attitude measurement greatly. Star trackers are one of the most accurate attitude measurement devices which can provide the attitude data as low as arc-seconds. Star image extraction is to process the star maps obtained by imaging stars in the sky and calculate the positions of star images. Its performances determine whether star trackers can measure the attitudes successfully. The noises in space degrade the star imaging of star trackers severely and bring more difficulties to star image extraction. In this paper, in order to withstand the noise interference, an anti-noise star image extraction algorithm based on YOLOv5 for star trackers was proposed which recognized the star images directly from the noisy star maps and lessened the difficulty of the star image extraction. Firstly, the working principle of star trackers was introduced. The star image extraction procedures were analyzed. The anti-noise star image extraction algorithm was proposed after the structure of YOLOv5 was introduced. Its procedures were described in detail. Secondly, the training and testing data sets were created by applying the developed star imaging simulation method to produce star images disturbed by Gaussian noise and Poisson noise. Then experiments were fulfilled. It proved that the proposed algorithm achieved a satisfactory recognition of the star images even with strong noises.