Robot Vision Group

National Laboratory of Pattern Recognition

Institute of Automation, Chinese Academy of Sciences

Automatic Celestial Spectra Classification and Recognition

(Supproted by Natural Science Foundation of China, National 863 Program of China, National Astronomical Observatories)


Scientific Goals

The Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) project as one of the National Major Scientific Projects undertaken by the Chinese Academy of Science. LAMOST is a quasi-meridian reflecting Schmidt telescope laid down on the ground with its optical axis fixed in the meridian plane. The aperture of LAMOST is 4m, enabling it to obtain the spectra of objects as faint as down to 20m.5 with an exposure of 1.5 hour. Its focal plane is 1.75m in diameter, corresponding to a 5° field of view, may accommodate as many as 4000 optical fibers. So the light from 4000 celestial objects will be led into a number of spectrographs simultaneously. Thus the telescope will be the one that possesses the highest spectrum acquiring rate in the world.

LAMOST adopts the active optics technique both for thin mirror and segmented mirror on the Schmidt corrector MA, as well as the parallel controllable fiber positioning system. With these new concepts and design, LAMOST is expected to be a unique astronomical instrument in combining a large clear aperture and wide field of view.

The engineering of LAMOST consists of eight subsystems, optic system, active optics and mirror supporting system, mounting and tracking system, telescope control system, focal plane instruments, telescope enclosure, observatory control and data processing, and input catalogue and survey strategy. The project will come into operation at the end of 2007.

The telescope will be located at the Xinglong Observing Station of National Astronomical Observatories, Chinese Academy of Sciences. As a national facility, LAMOST will be opened to the whole Chinese astronomical community. Along with the completion of the construction at the beginning of the 21st century, LAMOST will bring Chinese astronomy to a leading position in the large scale observations of optical spectra, and in the research field of wide field astronomy.

After the expected completion of the LAMOST project, voluminous spectra of celestial bodies will be collected. According to the design of the project, about 10,000 to 20,000 spectra will be collected in each observation night. Effective techniques and methods are needed urgently in order to deal with such voluminous data. Automatic classification approach and processing software system become quietly necessary. So our goal is to implement a software system integrating with main automatic spectral classification algorithms and some preprocessing methods.

Main Targets:

  1. To extract all the features of a spectrum;
  2. To classify and identify most of the spectrums from LAMOST automatic;
  3. To measure a spectrum automatic, including its red shift value, feature wavelengths, element abundance, etc.

Main Techniques

  1. Use PCA to construct models and classifying spaces;
  2. Use Morphology, wavelet, median filter to extract feature lines;
  3. Use Neural Network to construct classifiers;
  4. Use Rough Set to extract rules;
  5. Use Hough Transform, Pseudo-triangle to identify red shift values.

New Stars were Found!

We applied our method to a galaxy spectra dataset, which contains 294,843 galaxy spectra with high SNR (signal to noise ratio) from the SDSS-Data Release 7 (SDSS-DR7). Accordingly, we obtained 36 spectra of supernova candidates that include 12 known supernovae and 24 new supernova candidates as shown in Fig 1. Fifteen of the 24 new candidates have been reported, and the other nine are reported first. The results show that our method is feasible and effective. The primary advantage of our method is that it is not necessary to observe the sky area repeatedly for several times, and we search for supernova candidates just from the galaxy spectra obtained by LAMOST every night. Comparing with other large supernova surveys, the resource consumption of our method is almost negligible.


Fig. 1 Spectra of the 24 new supernova candidates we obtained. The top of each sub-figure is the original galaxy spectrum; the bottom is the spectrum without strong narrow lines(light gray) and smoothing results(dark gray)