Solar Difference Velocimetry Accuracy Analysis And Navigation Spectrum Selection For Deep Space Exploration

Deep space exploration,as an important development direction in the aerospace field,has higher requirements for the autonomous operation of spacecraft.Celestial navigation uses signals of natural celestial bodies as navigation targets,which can realize rapid positioning and attitude adjustment of spacecraft through three methods:angle measurement,range measurement and velocity measurement.Doppler velocimetry navigation is an emerging autonomous celestial navigation method in recent years.Using the spectrograph carried by the spacecraft to obtain the information of the optical spectrum lines,the radial velocity of the spacecraft relative to the navigation source can be caculated in real time.This method is highly autonomous,simple to solve,free from the limitation of ground support and the complexity of orbit dynamics model.In the current research on Doppler velocimetry,navigation sources can be divided into two categories: solar and extrasolar stars.As the only star in the solar system,the sun is closer to the earth than other stars and has strong spectral irradiance energy.However,the sun is so active that result in sunspots,flares and promineces,which will cause spectral changes,leading to the bias of velocity measurement.As for extrasolar stars,despite their large number,they are too far away from the earth,and most of the spectral lines with weak luminous flux are easily obscured by cosmic noise.Not all stellar spectra are suitable for velocimetry,so the method of screening and identifying the navigation source is the key to stellar spectral velocimetry.Based on the current research about Doppler velocimetry navigation,this paper discusses and studies the problems in solar spectrum and stellar spectrum velocimetry respectively,as follows:For solar spectrum velocimetry navigation,the velocity measurement accuracy of solar Doppler difference navigation is discussed and analyzed.Solar Doppler Difference navigation can reduce the influence of spectral distortion caused by solar activity on spectral redshift velocimetry,and plays a crucial role in the capture phase.However,there is still a lack of correlation analysis for possible errors in practical application.The accuracy of spectral velocimetry is an important performance index for the solar Doppler difference navigation.In order to analyze the influence of spectral variation on the accuracy of velocity measurement,this paper establish a velocimetry model based on the solar Doppler difference navigation,and propose a velocity accuracy analysis method.In this method,the reflected noise is filtered by wavelet transform,and then the high-resolution cross-correlation is used to solve the redshift value,so as to accurately calculate the velocity of the spacecraft.Combining the existing solar spectrum and Martian reflectance data,the 13 most obvious absorption lines are selected for velocity simulation and calculation.The experimental results show that the velocity accuracy of Doppler difference navigation can reach0.1312m/s.The width and depth of the spectral line,the planetary reflectivity,as well as the sunlight fluctuations and the reflected noise will all have a certain impact on the spectral velocimetry for the Solar Doppler difference Navigation.For stellar spectrum velocimetry navigation,a classification method based on deep learning is proposed to find the spectrum suitable for navigation.If the luminous flux of stellar spectrum is low and the spectral lines are not obvious,it is easy to be interfered by cosmic noise,covering up the information of the original spectral lines,and ultimately fails to provide effective velocity for celestial navigation for positioning and navigation.In order to provide a good navigation source target for the autonomous navigation of stellar spectral velocimetry,the spectral classification and recognition methods of a Convolutional Neural Network based on one-dimensional spectrum and a Residual Network based on folded spectrum are proposed.First,interpolation and interception are used to unify the resolution of the spectral data set,and then the maximum-minimum normalization and Borderline-SMOTE oversampling techniques are used to improve the spectral classification and recognition ability of the model.In view of the differences between the two kinds of neural networks,the processed one-dimensional spectral data of 1*4000 and the folded spectral data of64*64 are respectively used for training.The model uses accuracy,loss,confusion matrix and F1-score as evaluation indexes.The experimental results show that the residual network had more accurate classification ability,but at the cost of greater time cost.When it comes to spectral subclasses classification,it has high accuracy.According to the characteristics of spectral lines,A0-type spectrum is selected as the best navigation source,resulting in a fine identification.