聚类算法在天文学中的应用

发布时间：2018-06-23 01:49

本文选题：数据挖掘 + 聚类算法　；参考：《河北师范大学》2009年硕士论文

【摘要】： 随着大型光学望远镜的精度和深度不断提高,特别是巡天望远镜的发展,天文光学波段的数据急剧增加,成为探索各类天体和天文现象的物理本质的强有力武器。面对天文学“数据雪崩”和“信息爆炸”时代的到来,为了解决天文数据的复杂性、非线性、海量性和多波段性等一系列问题,借助数据挖掘技术来探索隐藏在数据中的有用信息势在必行。在此背景下,如何迅速准确地从这些海量数据中挖掘出所需要的信息,已成为天文学观测和数据处理的一个关键问题,直接影响着天文学发展和研究进程。本文利用一些适合于天文数据特点的数据挖掘技术和方法,主要是利用聚类算法对天体的自动分类和寻找出特殊天体的任务。工作主要包括以下三部分: (1)基于自动聚类算法的恒星/星系分类自动聚类算法(AutoClass)是基于贝叶斯模型确定最优类别的经典混合模型的非监督聚类方法。它对处理非线性和高维数据有很高的效率。应用AutoClass对SDSS巡天的恒星/星系进行自动分类,根据点源和展源的不同特征表现,选择使用了PSF(pointspread function,点扩散函数)星等和模型星等在五个波段的差值作为输入参数,并设置了合适的标准,从而获得合理的分类结果,对恒星和星系的分类正确率分别达到了99.51%和98.52%,表明AutoClass算法对此类数据聚类有很好的效率。 (2) SDSS巡天的恒星样本探索目的是探索SDSS恒星样本,找出其中的非恒星天体或特殊天体,从而建立一个纯的恒星样本。我们对SDSS巡天中具有光谱观测的恒星的测光数据应用AutoClass进行聚类分析,得到991个离群天体。然后将这些离群数据在NED和SIMBAD中认证,对于已证认的,发现它们绝大多数属于特殊天体,准确率高达90.7%,而对于未证认的,希望在更大口径更高精度的望远镜中证认,或许会发现一些有趣的天体或现象。 (3)星系形态分类星系是构成宇宙的组成单元,而形态是它的基本特性。星系形态的研究是理解星系物理属性的首要步骤。这里根据天文学的任务将星系分成早型星系和晚型星系两类。应用SDSS巡天的星系测光数据,基于五个星等和四个颜色的多组参数,参考不同的分类标准,研究了将AutoClass和k-means两种算法用于星系形态的自动分类。研究结果表明不论对AutoClass还是k-means而言,发现基于颜色的分类效果明显高于基于星等的效果;AutoClass和k-means的分类效率相当;与单纯地基于单参数的星系分类相比,自动化算法具有高效性、灵活性、能够处理高维数据等优点。在数据日益丰富的时代,自动化聚类算法将越来越显示出其优越性。
[Abstract]:With the continuous improvement of the precision and depth of large optical telescopes, especially the development of the Sky Survey Telescope, the data of astronomical optical band increase rapidly, and become a powerful weapon to explore the physical essence of various celestial bodies and astronomical phenomena. In the face of the advent of the age of "data avalanche" and "information explosion" in astronomy, in order to solve a series of problems such as complexity, nonlinearity, magnanimity and multi-band characteristics of astronomical data, It is imperative to explore useful information hidden in data by means of data mining technology. In this context, how to quickly and accurately extract the needed information from these massive data has become a key problem in astronomical observation and data processing, which directly affects the development and research process of astronomy. In this paper, some data mining techniques and methods suitable for astronomical data characteristics are used, mainly the automatic classification of celestial bodies and the task of finding special celestial bodies by clustering algorithm. The main work includes the following three parts: (1) the automatic star / galaxy classification clustering algorithm based on automatic clustering algorithm (AutoClass) is a classical algorithm based on Bayesian model to determine the optimal classification. Unsupervised clustering method for mixed models. It has high efficiency in dealing with nonlinear and high dimensional data. Using AutoClass to classify the stars / galaxies of SDSs, according to the different characteristics of point source and spread source, the difference between the PSF (pointspread function, point diffusion function and the model magnitude in five bands is selected as the input parameter. And set the appropriate criteria to obtain reasonable classification results, The classification accuracy of stars and galaxies is 99.51% and 98.52% respectively, which indicates that AutoClass algorithm is efficient for this kind of data clustering. (2) the purpose of star sample exploration in SDSS survey is: Explore SDSS stellar samples, A pure sample of stars is established by identifying the non-stellar objects or special objects. We use AutoClass to cluster the photometric data of stars with spectral observations during the SDSS survey and obtain 991 outliers. These outliers are then authenticated in NED and SIMBAD, and for those that have been identified, most of them are found to be special objects, and the accuracy is as high as 90.7, while for those that are not identified, they want to be identified in telescopes with a larger aperture and higher accuracy. Some interesting celestial bodies or phenomena may be found. (3) Galaxy morphological classification galaxies are the constituent units of the universe and form is its basic characteristic. The study of galactic morphology is the first step in understanding the physical properties of galaxies. Galaxies are divided into early and late galaxies according to astronomical tasks. In this paper, the automatic classification of galactic morphology is studied by using SDSS-survey data, based on the parameters of five star sizes and four colors, and referring to different classification criteria. The two algorithms, AutoClass and k-means, are applied to the automatic classification of galactic morphology. The results show that for both AutoClass and k-means, the classification effect based on color is significantly higher than that based on magnitude. The classification efficiency of AutoClass and k-means is similar to that of autoClass and k-means, and the automatic algorithm is more efficient than the classification based on single parameter. Flexibility, the ability to handle high-dimensional data and other advantages. In the era of increasingly abundant data, the automatic clustering algorithm will show more and more advantages.
【学位授予单位】：河北师范大学
【学位级别】：硕士
【学位授予年份】：2009
【分类号】：P152

【相似文献】