A Robust Method for High-Precision Celestial Positioning of Space Targets

The high-precision celestial positioning of space targets is constrained by star point centroid errors, star identification errors, and residual distortions in wide-field imaging. To improve the positioning accuracy and robustness under complex stellar-field conditions, this study focuses on improving star point centroid extraction and star identification. For star point centroid extraction, an improved effective point spread function (ePSF) fitting method is adopted to construct an ePSF model consistent with the actual imaging process, which characterizes the instrumental response, pixel sampling, and stellar intensity distribution, thereby improving the accuracy of sub-pixel centroid extraction. For star identification, a two-level matching method combining the inradius of star triangles and angular-distance constraints is proposed. Candidate screening, angular-distance constraints, and posterior validation based on a theoretical reference star map are used to reduce redundant matches and mismatching risks. Experiments on simulated star images show that the star identification success rate of the proposed method reaches 97.32%, outperforming traditional algorithms. In real star images, the star identification precision, star identification completeness, and F1 score are 93.59%, 90.14%, and 91.83%, respectively. When the 20-constant plate model is adopted, the average positioning errors of simulated and real star images are reduced to 0.86″ and 1.10″, respectively. Further increasing the model to 30 constants provides limited accuracy gain, which is insufficient to fully offset the cost of increased model complexity and parameter stability. The results show that the proposed method achieves a favorable balance among positioning accuracy, identification reliability, and model complexity.