We examine three approaches to the problem of source classification incatalogues. Our goal is to determine the confidence with which the elements inthese catalogues can be distinguished in populations on the basis of theirspectral energy distribution (SED). Our analysis is based on the projection ofthe measurements onto a comprehensive SED model of the main signals in theconsidered range of frequencies. We first first consider likelihood analysis,which half way between supervised and unsupervised methods. Next, weinvestigate an unsupervised clustering technique. Finally, we consider asupervised classifier based on Artificial Neural Networks. We illustrate theapproach and results using catalogues from various surveys. i.e., X-Rays(MCXC), optical (SDSS) and millimetric (Planck Sunyaev-Zeldovich (SZ)). We showthat the results from the statistical classifications of the three methods arein very good agreement with each others, although the supervised neuralnetwork-based classification shows better performances allowing the bestseparation into populations of reliable and unreliable sources in catalogues.The latest method was applied to the SZ sources detected by the Plancksatellite. It led to a classification assessing and thereby agreeing with thereliability assessment published in the Planck SZ catalogue. Our method couldeasily be applied to catalogues from future large survey such as SRG/eROSITAand Euclid.
展开▼
