07 - Systems, Subsystems and Equipment
AN IMPROVED END-TO-END STAR MAP RECOGNITION METHOD FOR AEROSPACE VEHICLE
Y. Zhang¹, Y. Yang², Q. Xiao², R. Li¹, Z. Xu¹; ¹Northwestern polytechnical University, China ;²China Academy of Launch Vehicle Technology, China
The fast speed and significant attitude maneuvers of high-dynamic Aerospace Vehicle in near space can lead to severe degradation of star images captured by the star sensor, resulting in large errors in starlight attitude determination. To address this issue, a fuzzy image-based end-to-end neural network recognition method is adopted. A main star mode segmentation algorithm is designed to obtain the approximate positions of stars, addressing the deficiencies of traditional star map recognition methods, in which the positions of star centroids are imprecise and cannot be identified. Subsequently, a densely connected local star map end-to-end restoration network is used to obtain the precise centroid positions of stars. Considering the extremely low signal-to-noise ratio and complex noise characteristics of real star maps, a fault-tolerant detection algorithm assisted by reference stars is proposed to eliminate misidentification and stars with excessively large positioning errors. Simulation results demonstrate that the Faster R-CNN method can accurately detect the main star modes, and the local star map restoration network can effectively restore degraded star maps. After restoration, the attitude accuracy can be maintained within [certain range], and the centroid positioning accuracy is improved compared to traditional methods. The methods proposed in this paper can effectively enhance the adaptability of starlight navigation systems in high-dynamic flight environments and have a certain corrective effect on the large initial deviations resulting from future spacecraft launches without reliance on external support.