基于高性能计算的LAMOST连续谱异常光谱的自动识别研究

发布时间：2017-12-26 22:06

本文关键词：基于高性能计算的LAMOST连续谱异常光谱的自动识别研究　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着LAMOST光谱观测工作的持续开展,光谱处理流程效率和准确率的逐步提高,拍摄的光谱质量也在不断提高。按照每个像素信噪比大于10作为检测光谱达标标准,达到要求的光谱占观测光谱的70%左右,而在正式发布的数据中也存在很多低质量光谱,在低质量光谱中占比例较大的就是连续谱异常光谱。连续谱异常是指恒星光谱在获得和处理过程中由于星际消光和流量定标等原因造成连续谱严重偏离黑体谱甚至中断的现象,这对光谱的谱线提取以及其他一些光谱研究工作带来负面影响,对连续谱异常光谱的研究亟待解决。LAMOST每个观测夜产生大约20,000-40,000条光谱数据,数据量级的不断增大使得对光谱处理流程自动化的要求越来越高,对异常光谱的处理也应达到更加高效的要求。本课题就是在此背景下提出来的,本文提出了一种基于距离度量的连续谱异常光谱的自动检测方法,主要工作有:(1)提取恒星光谱连续谱。分别提取待测光谱和对应类型模板光谱的连续谱,同时进行流量插值和归一化处理,使待测光谱和模板光谱处于相同的量纲下。(2)连续谱模板匹配。对待测光谱进行模板匹配,对(1)中提取的连续谱,在每个波长点计算待测光谱及其模板光谱的流量差值,分析流量差值的分布,检验有多少差值点分布在均值(β)附近的± a个标准差(δ)的范围内,进而确定是否有连续谱异常。(3)对待测光谱进行恒星光谱类型分类。由于大部分异常光谱的恒星类型未确定,而模板匹配的前提是已知待测光谱的恒星类型,本文采用测量模板光谱和待测光谱的Lick线指数之间欧式距离的方法,定义待测光谱类型为与其距离最近的模板光谱的类型。(4)在高性能平台下实现连续谱自动识别方法的并行计算。研究了Python在高性能平台的并行计算方法,通过调用MPI,将若干个FITS文件分给多个处理器核心,每个核心独立顺序执行,当最后一个核心计算完毕时,指定某个进程(一般是root)将各个核心的执行结果收集起来。总结来说,本文在高性能计算平台下实现的连续谱异常光谱自动识别方法,相比于单机计算达到了更加快速高效识别的效果。
[Abstract]:With the continuous development of the LAMOST spectral observation, the efficiency and accuracy of the spectral processing process are gradually improved, and the spectral quality of the photographed is also increasing. According to the signal to noise ratio greater than 10 of each pixel as the standard of the detection spectrum, the required spectrum accounts for about 70% of the observed spectrum, and there are also many low quality spectra in the officially released data. The larger proportion of the spectrum in the low quality spectrum is the continuous spectrum abnormal spectrum. Continuous spectrum anomaly refers to stellar spectra in obtaining and processing due to interstellar extinction and flow calibration caused by continuous spectrum of serious deviation from the blackbody spectrum even interrupted phenomenon, the spectral line extraction and the negative effects of some other research work on the spectrum, continuous spectrum spectral anomalies to be solved. LAMOST generates about 20000-40000 spectral data per observation night. The increasing data level makes the requirement of the automation of the spectral processing process more and more high, and the treatment of abnormal spectra should also achieve more efficient requirements. This topic is put forward under this background. In this paper, a distance spectrum based automatic detection method for abnormal spectrum of continuous spectrum is presented. The main works are as follows: (1) extraction of continuous spectrum of stars. The continuous spectrum of the measured spectrum and the corresponding type template spectrum is extracted respectively, and the flow interpolation and normalization processing are carried out at the same time, so that the measured spectrum and the template spectrum are in the same dimension. (2) continuous spectral template matching. The spectrum measured by template matching to treat, (1) continuous spectrum extraction, calculation of flow difference spectra and template spectra measured at each wavelength point, analysis of the distribution of the flow difference value, the number of test point distribution in mean difference (P) near the a + standard deviation (delta) range, and to determine whether there is a continuous spectrum anomaly. (3) to classify the spectral types of the stars with the spectrum. The most anomalous spectral types of stars is determined, and the premise of template matching is known to be measured spectral types of stars, this paper uses the method of Euclidean distance between the measured template spectra and measured the Lick line index is defined to be tested for its spectral type type template of the nearest distance. (4) the parallel computation of the method of automatic recognition of continuous spectrum is realized under the high performance platform. The study of the Python in high performance parallel computing platform, by calling MPI, a plurality of FITS files to multiple processor cores, each core independent sequence, when the last core to complete the calculation, specify a process (usually root) the results of the implementation of the core collection. In conclusion, the continuous spectrum automatic recognition method based on high performance computing platform achieves faster and more efficient recognition than single machine computing.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P144.1;TP38

【相似文献】