By virtue of the sensitivity of the XMM-Newton and Chandra X-ray telescopes, astronomers are capable of probing increasingly faint X-ray sources in the universe. On the other hand, we have to face a tremendous amount of X-ray imaging data collected by these observatories. We developed an efficient framework to classify astronomical X-ray sources through natural grouping of their reduced dimensionality profiles, which can faithfully represent the high dimensional spectral information. X-ray imaging spectral extraction techniques, which use standard astronomical software (e.g., SAS, FTOOLS and CIAO), provide an efficient means to investigate multiple X-ray sources in one or more observations at the same time. After applying independent component analysis (ICA), the high-dimensional spectra can be expressed by reduced dimensionality profiles in an independent space. An infrared spectral data set obtained for the stars in the Large Magellanic Cloud,observed by the Spitzer Space Telescope Infrared Spectrograph, has been used to test the unsupervised classification algorithms. The least classification error is achieved by the hierarchical clustering algorithm with the average linkage of the data, in which each spectrum is scaled by its maximum amplitude. Then we applied a similar hierarchical clustering algorithm based on ICA to a deep XMM-Newton X-ray observation of the field of the eruptive young star V1647 Ori. Our classification method establishes that V1647 Ori is a spectrally distinct X-ray source in this field. Finally, we classified the Xray sources in the central field of a large survey, the Subaru/XMM-Newton deep survey, which contains a large population of high-redshift extragalactic sources. A small group of sources with maximum spectral peak above 1 keV are easily picked out from the spectral data set, and these sources appear to be associated with active galaxies. In general, these experiments confirm that our classification framework is an efficient X-ray imaging spectral analysis tool that gives astronomers insight into the fundamental physicalmechanisms responsible for X-ray emission and, furthermore, can be applied to a wide range of the electromagnetic spectrum.
Library of Congress Subject Headings
Astronomical spectroscopy--Data processing; Source separation (Signal processing)--Statistical methods; Multivariate analysis
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Mu, Bo, "Unsupervised spectral classification of astronomical x-ray sources based on independent component analysis" (2007). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus