Research On Star Map Recognition Algorithm Based On Machine Learning Perspective

With the development of the times,China’s space technology has made a series of new advances,and the requirements for accuracy and precision of spacecraft attitude adjustment have become higher and higher.The star sensor is the best comprehensive attitude sensor,and star map recognition is one of the key technologies of the star sensor.Star map recognition refers to the process of constructing a navigation star database,entering the selected navigation star features into the navigation star database for storage,then the star sensor obtains the star map by photographing,extracts the observation star features according to the star map,and compares the observation star features with the features stored in the navigation star database to calculate the position of the observation star in the celestial coordinate system,so as to determine the attitude of the spacecraft.The study of the key technologies in the process of star map identification is of great importance for the development of star-sensitive navigation technology.In this paper,we study the establishment of a navigation star library,the simulation of actual star charts,the processing of star chart images and the recognition and matching of star charts from the perspective of the star chart recognition algorithm as a whole.This paper firstly introduces the basic knowledge involved in the star chart recognition algorithm,including the stellar theory,star-sensitive instrument theory and the basic theory of neural networks.Secondly,the SAO catalogues were screened,duplicate stars,stars with missing magnitudes,double stars and variable stars were eliminated in order,and stars with magnitudes higher than 6.0 were eliminated,and 4 988 stars were screened.Finally,3 430 stars were screened through appropriate processing of the lower angular pitch pairs and homogenisation of the screened stars,and a navigation star library was successfully built.According to the results of the goodness-of-fit test,the library of navigation stars is uniformly distributed on the sphere,although there are still a few regions where stars are clustered.Then,we simulate the actual star map with the navigation star library,and use global thresholding,mean filtering and Gaussian filtering to reduce the noise of the star map,and compare the advantages and disadvantages of each noise reduction algorithm with the signal-to-noise ratio and peak signal-to-noise ratio.Next,the simulated star maps are detected in the connected domain,and the square-weighted center-of-mass method is used to extract the center-of-mass of the star points in the star maps.Finally,the triangle algorithm and the raster algorithm are used to extract the features of the star map,where the triangle algorithm contains 1 107 259 feature triangles and the raster algorithm contains only 3 430 sparse feature matrices.In this paper,the neural network algorithm is combined with the raster algorithm to propose a new star map recognition algorithm,which can achieve a recognition rate of 95% under simulated conditions.The new method is more resistant to interference and robust relative to the triangle algorithm.The conclusion of this paper is that the neural network algorithm-based star map recognition method has higher recognition rate and recognition efficiency than the traditional method,and is more resistant to interference,and the neural network model has greater application value in the field of star map recognition.