Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.atnf.csiro.au/research/LVmeeting/posters/MWhiting-poster.pdf
Дата изменения: Fri Jul 6 05:59:19 2007
Дата индексирования: Sun Dec 23 09:43:42 2007
Кодировка:
Astronomers!
Do you know where your galaxies are?
Introducing the Duchamp Source Finder
Matthew W hiting, Australia Telescope National Facility, CSIRO

The Problem: Making the most of your surveys
Large-scale spectral-line radio surveys are an effective way to study populations of galaxies, masers, or sites of star-formation. Surveying on a large scale allows one to both probe to the fainter flux levels and remain sensitive to the rare but interesting objects at the brightest fluxes. Effective use of large-scale surveys is dictated by the efficiency of detection of the objects of interest. Finding the extreme bright objects is often the easy bit ­ to find the more typical sources you need to push down to fainter fluxes close to the noise level. The complexity and volume of three-dimensional survey data also requires a large degree of automation and reliability to find the desired sources quickly and in a uniform way.

The Solution: The Duchamp Source Finder
Duchamp is our solution to the problem of three-dimensional source finding. It is a stand-alone software package, designed to work with spectral-line FITS cubes and produce source lists and graphical results showing detected objects. Duchamp is optimised for the case of a large number of separated sources embedded in a cube dominated by noise ­ the typical situation expected for HI or maser surveys.
Figure 2: Example graphical output of Duchamp showing the zeroth moment of each detection, in its appropriate spatial location. This is provided as a postscript file and displayed in a PGPLOT window.

Innovations in Searching
A key aspect of source detection is the application of a threshold. This is calculated by using either a simple n-sigma cutoff, or derived through the use of the False Discovery Rate method1,which controls the number of false detections. The statistics for the threshold calculations are obtained through robust methods, and can be measured from the full cube or a specified subset. Searching in 3D is in general a complex problem. Duchamp approaches this by searching in each 2D channel map separately2, and comparing detections made in one channel with those in neighbouring channels. An efficient merging algorithm then merges the 2D detections to form 3D objects.
Figure 3: Example graphical output showing the spectral nature, the zeroth moment and the basic properties of a detection. A postscript file is produced showing such a plot for each object.

Beating the noise
The limiting factor in detecting faint objects is the background noise, and Duchamp provides ways to minimise its effects. Simple smoothing before searching is possible, either spectrally using a Hanning filter, or spatially using a 2D Gaussian kernel. Alternatively, the cube can be reconstructed using the Ю trous wavelet technique3. This filters the cube at a range of scales, thresholding each scale and only keeping those pixels with significant signal. This very effectively removes noise from the cube, allowing searching to be done on much cleaner data.
Figure 4: Typical output to screen at completion of a Duchamp run, showing the range of parameters calculated. Other possible outputs include formats f or use with Karma software or Virtual Observ atory applications.

Obtaining Duchamp and Further Development
Duchamp is easy to install and run, and is available from http://www.atnf.csiro.au/people/Matthew.Whiting/Duchamp Any and all feedback is welcome! Duchamp will continue to be developed, particularly in response to users' requirements. It will also form the basis for analysis software under development for use with the Australian SKA Pathfinder.

Figure 1: An example of the Ю trous algorithm used in one dimension to reconstruct a spectrum. Note the doublehorned shape of the resultant main object, and the preserv ation of the weak f eatures around channels 170 & 900.

References
1 2 3

Acknowledgements The n am e Ducha mp com es from the renowned Cub ist & Dad aist artist M arc el Duchamp, who p ion eered th e art of th e read ym ad es, o r `found objects'. The logo at top right is appropriated from Duchamp's 1913 read ym ade `Bicycle Wheel'.

M iller et al 2001, AJ 122, 3492. Using th e techn ique of Lutz 1980, Th e Computer Journ al 23 , 262 Starck et al 1997, ApJ 482, 1011 is a good examp le of th e techn ique.

contact: Matthew W hiting phone: +61 2 9372 4683 email: Matthew.W hiting@csiro.au web: www.atnf .csiro.au/people/Matthew.W hiting/Duchamp

Duchamp m akes use o f the PGPLOT, CF ITS IO and W CSLIB libraries. For more information, see th e U ser's Gu id e, or the fo rthcom in g pap er.