The CNN classification of galaxies by their image morphological peculiarities. Issue Volume 16:Issue S362(2020) (June 2020)

• Record Type: Journal Article
• Authors: Dobrycheva, D.; Khramtsov, V.; Vasylenko, M.; Vavilova, I.
• Editors: Bisikalo, Dmitry; Wiebe, Dmitri; Boily, Christian
• Is Part Of: Proceedings of the International Astronomical Union. Volume 16:Issue S362(2020)
• Page Start: 111
• Page End: 115
• Publication Date: 2020-06
• Subjects: methods: data analysis -- galaxies: general -- surveys -- methods: convolutional neural networks, etc.; Astronomy -- Congresses; Astronomy -- Periodicals; 520
• DOI: 10.1017/S1743921322001259
• Languages: English
• ISSNs: 1743-9213

Abstract

Abstract: Multidimensional mathematical analysis, like Machine Learning techniques, determines the different features of objects, which is difficult for the human mind. We create a machine learning model to predict galaxies' detailed morphology (∼ 300000 SDSS-galaxies with z < 0.1) and train it on a labeled dataset defined within the Galaxy Zoo 2 (GZ2). We use convolutional neural networks (CNNs) to classify the galaxies into five visual types (completely rounded, rounded in-between, smooth cigar-shaped, edge-on, and spiral) and 34 morphological classes attaining >94% of accuracy for five-class morphology prediction except for the cigar-shaped (∼ 87%) galaxies.