Research And Application Of Data Processing Technology On LAMOST Survey Spectra

The advent of large survey telescopes has brought huge amounts of data to the field of astronomy.Spectroscopic surveys,represented by the Sloan digital sky survey(SDSS)in US and Guo Shou Jin telescope(Large sky Area Multi-Object fiber Spectroscopic Telescope,the LAMOST)independently developed by China,make a rapid expansion of information acquired through spectroscopy.Analyzing the obtained celestial spectra,we can get the physical,chemical and kinematic properties of the objects,and even can limit their evolution process and so on.LAMOST telescope is a multi-objective spectral telescope with the largest field of view and the highest spectral acquisition rate in the international astronomical community.It is the first to realize the large-scale sky survey of observing thousands of celestial spectra at the same time.Since 2011,LAMOST has completed 6 years of large-scale sky survey observation,acquired the largest stellar parameter catalog in the world,and stored more than 9 million spectra of DR5 data set.The astronomical big data obtained by LAMOST enriches the global astronomical observation database.Based on this valuable data,our country's astronomy research in some fields,especially the study of the Milky Way galaxy,got rid of the situation of relying on foreign telescopes,and data enhanced the overall strength of our country's astronomy research.In addition,data-driven astronomical research has gradually become an important field of astronomical research.The acquisition of surveyed data,data processing and analysis platform,statistics and depth learning methods are studied in this paper.Focal ratio degradation is found in the optical system.Based on the practical LAMOST observation,the reliability of the data is analyzed from first stage of its producing.For the tens of thousands of A-type star spectral data of LAMOST,R language is introduced to analyze the massive spectral data,using the existing statistical tools to achieve efficient data processing.The systematic analysis of the difference between the LAMOST classification results and the template is made.A set of templates based on LAMOST observation data is established to analyze the B-type star spectra in the massive spectral data set of LAMOST,which have low observation ratio but are of great research value.The creative work in this paper includes:Firstly,in the optical system,the focal ratio degradation of the LAMOST large-core optical fibers is studied based on the shape of the output spot.By combining with the characteristics of CCD spectral image,the method of contour fitting is used to transform the two-dimensional CCD spectral image into one-dimensional spectrum,simulate the red-end data of LAMOST,and select the sampling points in the range of 5000?-6000?.The spectral line of this band(including Fe emission)is convoluted with the obtained annular speckle profile.The effect of focal ratio variation on the peak value of the Fe emission line is studied.The correctness of eliminating the deviation caused by annular speckle by hanging flat-field diffuse reflector screen in front of the main mirror Mb during the official sky survey observation is clarified,and the analytical ability of the astronomical spectral data is compacted to ensure the rationality.Secondly,R language platform are introduced to realize the data read and visualization by using the international standard format of the astronomical community data FITS load RFITSIO packages.R language packages can well mine and extract the main characteristic of spectral samples,to efficiently obtain information hidden in astronomical spectra.The statistical method integrated in R language is fully used to complete the pattern recognition and data mining of astronomical spectra,which improves the efficiency of data analysis.The R language based tools we used in this work include data fitting: data modeling,extreme value method,least square method,maximum likelihood method,nonlinear regression;Data interpolation: interpolation and heterodyne,spline method;Data smoothing: moving average,spectrum analysis;Distance classification: Euclidean distance,Markov distance;Correlation analysis: autocorrelation,cross correlation,delay,image translation;Dimension reduction of high-dimensional data: PCA,etc.The above methods are used to extract information from astronomical data,which provides a basis for the interpretation of spectral line characteristics of and the exploration of massive rare celestial bodies or unknown species.We also grouped LAMOST released F,G,K stellar spectra in to bins according to Kurucz model.After normalization,we measured more than 20 thousand spectra of the LAMOST,and compared with Kurucz template library,the results showed that the accuracy and reliability of the physical parameters measured by LAMOST pipeline is high,which provide a basis of optimization for popular atmospheric models.Thirdly,the Lick line index in astronomical spectrum is used to extract stellar features effectively.Multivariate linear regression and random forest algorithm were used to estimate the effective temperature of A-type stars with line index using Astrostat clustering analysis.Within the range of random forest training set data,regression prediction solves the problem of model over fitting,and the strong correlation between data is analyzed and explained by comparing the results of two kinds of regression.Ridge regression method was used to solve the problem of least square regression instability,so as to effectively use line index to predict the effective temperature of A-type stars,and obtain the method of correctly predicting the effective temperature of spectral data from the multidimensional parameter space of LAMOST large sample.Finally,the supervised clustering is performed according to all measured B-type stellar spectra released by LAMOST DR5,and the majority are marked as B6(7662)and B9(3969).Then selecting the B-type template from ELODIE spectrum library analyzed the intra-class distance of each sub-type through linear discriminant.According to the empirical template,a set of the new classification template of subtype for B-type spectra is successfully constructed,as an important supplement to the classification template library of LAMOST.