Precision improvement method of star centroid positioning based on multi-image super-resolution reconstruction for fine guide sensor
The accuracy of the fine guide sensor's star centroid positioning determines the accuracy of the visual axis attitude calculation of the space telescope.To improve the positioning accuracy of the star centroid of the fine guide sensor,a star image super-resolution reconstruction method based on the deep wavelet recurrent neural network is proposed.Firstly,the micro-scanning technology is used to obtain the sub-pixel misalignment low-resolution star image sequence,and the wavelet domain features of the low-resolution star image are extracted by using the wavelet encoder while the noise of the low-resolution star image is constrained by the wavelet coefficients.The registration process of the input star image sequence is integrated into the network learning.Secondly,the convolutional gate recurrent neural unit is used to fuse the features of the extracted star image sequence.Finally,the inverse wavelet decoder is utilized to decode the wavelet domain features output by the multi-feature fusion module.In this way,the de-noising and super-resolution reconstruction based on low-resolution star image sequences are realized.The experimental results show that the square-weighted centroid method is used to obtain the centroid positions of each star point in the original star image and the reconstructed star image with super-resolution.Compared with the former,the average centroid positioning accuracy and stability of each star point in the X direction are improved by 64.76％and 19.15％,respectively.In the Y direction,the accuracy and stability are improved by 75.35％and 26.14％,respectively.
fine guide sensorstar point centroid positioningsuper-resolution reconstructionwavelet signal processingconvolutional gate recurrent neural networks
NETL
NSTL
万方数据
