Unsupervised classification of SDSS galaxy spectra

Context. Defining templates of galaxy spectra is useful to quickly characterise new observations and organise databases from surveys. These are usually built a pre-defined classification based on other criteria. Aims. We present an unsupervised 702 248 galaxies quasars with redshifts smaller than 0.25 that were retrieved the Sloan Digital Sky Survey (SDSS) database, release 7. Methods. The first corrected for redshift, then wavelet-filtered reduce noise, finally binned obtain about 1437 wavelengths per spectrum. clustering algorithm Fisher-EM, relying discriminative latent mixture model, was applied these spectra. full set several subsets 100 000 300 analysed. Results. optimum number classes given by penalised likelihood criterion 86 classes, which 37 most populated gather 99% sample. established subset 302 214 Using cross-validation techniques we find this agrees results obtained average misclassification error 15%. large very small tends increase rate. In paper, do initial quick comparison our literature templates. Conclusions. This time automatic, objective robust such mean can be used as majority in Universe.