The Application Of Clustering Algorithms In Astronomy

With the precision and the depth increasing of big optical telescope,especially the development of sky survey telescope,data of astromomical optical bandpass rapid raise,and them has become forceful weapon to explore the physical essence of all kinds of celestial body and astromomical phenomenon.With the coming era of"data avalanche" and "data explosion" in astronomy,in order to solve a series of problems of complexity,nonlinearity, mass and multiband about astronomic data,it is necessary to explore the potential and useful information hidden in the data by means of data mining technologies.In this paper,we make use of some data mining technologies and methods meeting the characteristics of astronomical data,mainly apply the clustering algorithms to automatically classify celestial objects and search the peculiar celestial objects.The main contributions are following:(1) Classification of Stars/Galaxies Based on AutoclassAutoclass is an unsupervised method of classical mixed model that is based on Bayesian model to determine the optimal classes.It has higher efficiency to deal with the nonlinear and high-dimensional data.Applying AutoClass to automatically classify stars/galaxies according to the different performance of pointed sources and extended source,we use the difference value of PSF magnitude and model magnitude in five wavebands as the input parameters,and set suitable criterion,then obtain a reasonable classified result.The classification accuracy about stars and galaxies add up to 99.51%and 98.52,respectively,which shows that AutoClass algorithm has a better efficiency to deal with these data.(2) Exploration of SDSS Stellar DatabaseThe purpose is to explore SDSS stellar sample,to find the peculiar objects and construct a pure stellar sample.We apply AutoClass to analyze stellar photometric data with spectra and obtain 991 outliers.We identify theses outliers by NED and SIMBAD,and find the most of the identified outliers belong to the peculiar objects,the accuracy adds up to 90.7%.It is hoped that the unidentified objects will be follow-up observed by a larger and better telescope, some new interesting objects or phenomena will be detected.(3) Automated Morphological Classification of GalaxiesGalaxy is the elementary cell of constituting the Universe,and one of the basic characteristics of galaxy is its morphological type.Morphology study of galaxies is the first step to understand the galactic physics.We classify the galaxies into early-type galaxies and late-type galaxies according to the task of astronomy here.Taking many sets of parameters based on five magnitudes and four colors as input pattern,respectively,given different classification criterions,we utilize AutoClass and k-means to classify galaxies into early- and late-type galaxies with SDSS galactic photometric data.The experimental result shows that no matter AutoClass or k-means,the performance based on colors is superior to that based on magnitudes;the accuracy of AutoClass and k-means is comparable;compared to the classification only based on one physical parameter,the automated algorithms has many merits of effctiveness,flexibility,ability of dealing with high dimensional data.In the era of the more abundant data,automated clustering algorithms show more superiorities.