Prediction-selection-extraction: star point extraction from daytime star maps based on the randomness of stellar imaging energy.

Daytime star detection represents a significant advancement over traditional methods, with applications in astronomical navigation, atmospheric inversion, and satellite-ground communication. However, daylight conditions impose challenges such as limited exposure time, elevated background noise, and pronounced atmospheric turbulence. These factors reduce the accuracy, success rates, and adaptability of traditional star point extraction algorithms, directly affecting the performance of attitude and orientation systems. In this study, the key factors affecting high-precision star point extraction from daytime star maps are analyzed and what we believe to be a novel method called prediction-selection-extraction (PSE) for star point extraction based on the randomness of stellar imaging energy is developed. The PSE method consists of three stages: 1) prediction, where prior information from a star map is used to predict the approximate positions of star points; 2) selection, where multiple star map frames are selected adaptively based on the characteristics of atmospheric turbulence; 3) extraction, where a multi-frame joint adjustment model based on stellar motion parameters is used to extract the centroids of star points with high precision. Unlike traditional algorithms based on image processing, PSE fully leverages the prior information in star maps and effectively avoids the adverse effects of the two assumptions mentioned above. Experimental results demonstrate that PSE performs well on daytime star maps under various complex conditions, achieving an extraction success rate of over 97%. The extraction accuracy of the PSE algorithm reaches 0.07 pixels-271% higher than that of traditional single-frame extraction algorithms-thus meeting the demands of various engineering applications.