Research On Intelligent Star Image Processing Method Of Micro-star Tracker In Complex Space Environment

In this paper,the intelligent star image processing of micro star tracker is studied.The research is centered on two key issues of star image clutter suppression and distortion correction for miniaturized star sensor,and a digital dynamic star simulator system framework for high fidelity star image simulation is developed.The micro-star tracker is susceptible to the influence of space environment when it is in orbit.The background of the star image changes dramatically when the shield design of the star tracker is imperfect or various types of reflected light enter the field of view(FOV).In addition,when the micro star tracker is running in orbit,a large number of non-linear distortion of the optical system is caused by the error of optical lens processing,optical path assembly,vibration,stress,etc.The maximum distortion magnitude can be more than two pixels,which seriously affects the positioning accuracy of the star tracker.Star images are subject to strong noise caused by complex spatial environmental impacts.In this paper,the complex spatial environment of the star tracker is analyzed and modeled,the imaging characteristics of different stray light interference in the star image are summarized,and a single star image strong stray light interference removal method based on SLIC-DBSCAN is proposed.The method first enhances the contrast between background noise and strong interference based on the constrained LC algorithm,then uses the super-pixel segmentation algorithm to segment the salient image,and finally extracts the multidimensional features from each super-pixel,and combines the super-pixels with different features using the density-based clustering algorithm.Applying this method to real on-orbit image can effectively remove strong stray light in the image.To solve the problem of star image distortion,a single-layer two-dimensional Legendre neural network(2DLNN)is presented,which can automatically correct geometric distortion of star tracker.Offline training algorithm based on batch input star image and on-orbit training algorithm based on sequential input star image are designed,respectively.The 2DLNN network realizes the ground and on-orbit correction of geometric distortion of star tracker optical system.It has the characteristics of self-learning and lifelong learning.The single layer network has simple structure,fast convergence,and is suitable for on-orbit implementation.The optimal parameters of the polynomial are determined by experiments.The simulation results show that the algorithm convergence can be completed in 2,000 to 2,500 seconds with online sequential recursion in the loworbit satellite ground-to-ground orientation mode,the star distortion magnitude can be reduced to 0.04 pixels,and the single frame training time is less than 0.6ms,which is suitable for different types of distortion.The positioning accuracy of the star tracker is greatly improved and the design cost of the star sensor is reduced.At the end of the paper,a high fidelity star image generation strategy is proposed for the star simulator which cannot directly simulate the state of the star tracker in orbit.The digital dynamic star simulator system is completed by combining the ideal digital star image with the simulated spatial environment.Spatial environment scenarios such as transient effects of space particles,earth albedo,spacecraft trajectory and space debris albedo are mainly simulated.Star image on-orbit imaging is simulated,which lays a solid foundation for verifying star tracker related algorithms and testing the performance of star tracker on-orbit operation.