Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
EP: Fachverband Extraterrestrische Physik
EP 6: Postersitzung
EP 6.3: Poster
Mittwoch, 11. März 2015, 16:45–18:45, Foyer Ebene G.10
Automatic classification of box and peanut shaped bulges using self organizing maps — Baris Özcan1, •Rainer Lütticke1, and Kai Lars Polsterer2 — 1Hochschule Bochum, Lennershofstr. 140 — 2Heidelberger Institut für Theoretische Studien
There are several large studies to classify the morphology of bulges, e.g. peanut or box shaped (b/p) (Yoshino & Yamauchi, 2015, MNRAS 446, 3749; Lütticke, Dettmar, & Pohlen, 2000, A&AS 145, 405 [LDP]). Both referenced classifications are visual and very time consuming. An objective classification of bulge types would be useful. Therefore we developed software for an automatic classification of bulge types using unsupervised machine learning methods (self organizing maps, a kind of artificial neural network). This software is based on Polsterer, Gieseke, & Kramer (2012, ASPCS 461, 561). As input, our software gets preprocessed images (120x80 pixels). The preprocessing includes alignment with the major axis, scaling the galaxy image to the size of the bulge, and cutting out the region of the bulge. We give the image pixels in the region of the b/p structure a higher weighting in the classifying algorithm. Our sample includes 88 galaxies belonging to SDSS and fulfilling criteria for inclination, size, and Hubble type. With a set of 44 images the map (16 neurons = 16 classes) is trained and then the other set is classified using the trained map. However, the bulge types as defined in LDP are not well reproduced by these classes and automatically classified bulges are often wrong classified in comparison to LDP. We conclude that our approach has to be optimized or automatical classification for bulge types is not possible at all.