Automatic Morphological Classification of Galaxies: Convolutional Autoencoder and Bagging-based Multiclustering Model

In order to obtain morphological information of unlabeled galaxies, we present an unsupervised machine-learning (UML) method for classification which can be summarized as two aspects: (1) the methodology convolutional autoencoder (CAE) is used reduce dimensions and extract features from imaging data; (2) bagging-based multiclustering model proposed classifications with high confidence at cost rejecting disputed sources that are inconsistently voted. We apply this on sample galaxies $H<24.5$ in CANDELS. Galaxies clustered into 100 groups, each contains analogous characteristics. To explore robustness classifications, merge groups five categories by visual verification, including spheroid, early-type disk, late-type irregular, unclassifiable. After eliminating unclassifiable category inconsistent voting, purity remaining four subclasses significantly improved. Massive ($M_*>10^{10}M_\odot$) selected investigate connection other physical properties. The scheme separates well U-V V-J color space Gini-$M_{20}$ space. gradual tendency S\'{e}rsic indexes effective radii shown spheroid subclass irregular subclass. It suggests combination CAE multi-clustering strategy cluster similar yield high-quality classifications. Our study demonstrates feasibility UML analysis would develop serve future observations made China Space Station telescope.