基于残差分布度量的LAMOST M矮星细分类研究

发布时间：2018-10-17 19:12

【摘要】：随着天文观测仪器设备以及数据收集能力的大幅度提高,大量的巡天项目会产生海量的巡天数据,如我国的LAMOST巡天项目。如何对这些海量的光谱数据进行有效的处理与分析,是天文学研究中一个重要的研究内容。通过对光谱数据进行分析处理,可以确定恒星大气物理参数和天体的频谱空间分布等形态结构信息。恒星光谱分类是天文数据处理中一项非常重要的工作,主要应用于大量的巡天数据搜寻已知类型的天体光谱。通过事先给出有类别的两类或多类光谱训练一个分类器,对于新的光谱通过该分类器给出一个分类。模板库可以看作为一个训练好的分类器,将待测光谱与模板库中的光谱进行匹配以获得分类结果。本文就是利用残差分布度量的方法对LAMOST M矮星光谱进行细分类研究。残差分布度量是一种光谱间的距离度量方法,计算光谱之间的距离时,先将两条光谱进行归一化处理,之后计算对应波长采样点处的残差,最终以残差分布的标准差作为光谱之间的距离。对待测光谱进行分类时,首先需要对待测光谱进行插值、光谱归一化等预处理操作；之后利用残差分布距离度量方法计算出待测光谱与模板库中每条模板光谱之间的距离,从中找出距离最小的那一条模板光谱；最终将该模板光谱的类型作为待测光谱的类型。本文分别以模拟光谱和实测光谱作为实验数据进行了分类实验,其中模拟光谱由模板光谱生成,实测光谱采用的是LAMOST DR2中的M星实测光谱。实验结果表明,残差分布度量方法能比较准确地对光谱数据进行分类,与传统分类方法相比分类效果更好。光谱的分类效果受信噪比、离群点以及残差标准化系数等因素的影响。随着信噪比的增大,光谱分类效果越来越好。总体来说,利用残差分布度量方法对天体光谱进行分类是具有可行性的。本文中的分类方法和处理流程有待于进一步的研究,使其达到一个更好的分类效果,从而使该方法得到更加广泛地应用。
[Abstract]:With the great improvement of astronomical observation instruments and data collection ability, a large number of sky survey projects will produce massive data, such as the LAMOST survey project in China. How to effectively process and analyze these massive spectral data is an important research content in astronomical research. By analyzing and processing the spectral data, the morphological structure information such as the physical parameters of the stellar atmosphere and the spatial distribution of the celestial spectrum can be determined. Star spectral classification is a very important work in astronomical data processing. A classifier is trained by giving two or more kinds of spectra with a class in advance, and a classification of the new spectrum is given by the classifier. The template library can be regarded as a trained classifier. The spectrum to be tested is matched with the spectrum in the template library to obtain the classification results. In this paper, the method of residual distribution metric is used to subdivide the spectrum of LAMOST M dwarf star. The residual distribution measure is a method of measuring the distance between spectra. When calculating the distance between spectra, the two spectra are normalized first, and then the residuals at the sampling points of corresponding wavelength are calculated. Finally, the standard deviation of the residual error distribution is taken as the distance between the spectra. When classifying the measured spectrum, the interpolation and normalization of the measured spectrum are needed first, and then the distance between the measured spectrum and each template spectrum in the template library is calculated by using the residual distribution distance measurement method. Finally, the type of the template spectrum is taken as the type of the spectrum to be measured. In this paper, the simulated spectra and the measured spectra are used as the experimental data, respectively. The simulated spectra are generated from the template spectra, and the measured spectra are the M-star measured spectra in LAMOST DR2. The experimental results show that the residual distribution measurement method can classify the spectral data accurately, and the classification effect is better than the traditional classification method. The effect of spectral classification is affected by SNR, outlier and residual standardization coefficient. With the increase of SNR, the effect of spectral classification is getting better and better. In general, it is feasible to classify celestial spectra by residual distribution measurement. The classification method and processing flow in this paper need to be further studied to achieve a better classification effect, so that the method can be applied more widely.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P144;P111

【共引文献】