Facing with the requirement of automated processing and analysis for the LAMOST spectral data, this thesis focuses on automated searching for specialobjects–supernovae. The main contributions of this thesis include following four aspects: 1 We propose two kinds of methods to predict the redshift of galaxy spectra automatically. One is an integrated approach.Galaxy spectra are classified into into spectra of active galaxies and normal galaxies firstly based on Fisher Discriminant Analysis, then the density estimation technique is used to predict redshift for the active galaxies, and the cross-correlation technique is used to predict redshift for the normal galaxies. Experiments show that the correct rate of redshift prediction is higher than other three methods. The other is an improved method based on the similarity measure. Because spectral lines cannot usually be extracted accurately, in the method we use a number set varying with scales as redshift candidates, instead of determining redshift candidates according to the extracted spectral lines. Considering the relationship between the redshift measurement accuracy and spectral resolution, we perform the inner interpolation for each spectrum to be measured. For real spectral data from Sloan Digital Sky Survey (SDSS), the test results show that the correct rate of redshift measurement can be raised to up to 99%. 2 We propose a new automated method to decrease sample volume, whichcan be successfully used to reduce the range of searching for Ia supernova (SN) candidates in large number of galaxy spectra according to the rare attribute of supernova. Through the PCA analysis for Nugent’s simulation supernova template library, we construct a supernova spectral eigen-space and define a supernova statistical characterization vector (SNSCV) for each galaxy spectrum. Then based on Local Outlier Factor (LOF), we propose a method to automatically reduce the range of searching for supernova candidates. In addition, according to the same idea, we propose a method to further reduce the range of searching for supernova candidates using similarity measure instead of Euclidean distance. Because spectral data are usually high dimensional, Euclidean distance sometimes can not adequately reflect the similarity between the two spectra, thus we give the definition of local outlier factor based on similarity (SBLOF), and prove that the similarity between the two reconstructed spectra is equal to the similarity between th...
修改评论