Affiliation:
1. ICRAR M468, The University of Western Australia, 35 Stirling Hwy, Crawley, WA 6009, Australia
2. Research School of Astronomy and Astrophysics (RSAA), Australian National University, Canberra, ACT 2611, Australia
Abstract
ABSTRACT
Classifying the morphologies of galaxies is an important step in understanding their physical properties and evolutionary histories. The advent of large-scale surveys has hastened the need to develop techniques for automated morphological classification. We train and test several convolutional neural network (CNN) architectures to classify the morphologies of galaxies in both a 3-class (elliptical, lenticular, and spiral) and a 4-class (+irregular/miscellaneous) schema with a data set of 14 034 visually classified SDSS images. We develop a new CNN architecture that outperforms existing models in both 3-way and 4-way classifications, with overall classification accuracies of 83 and 81 per cent, respectively. We also compare the accuracies of 2-way/binary classifications between all four classes, showing that ellipticals and spirals are most easily distinguished (>98 per cent accuracy), while spirals and irregulars are hardest to differentiate (78 per cent accuracy). Through an analysis of all classified samples, we find tentative evidence that misclassifications are physically meaningful, with lenticulars misclassified as ellipticals tending to be more massive, among other trends. We further combine our binary CNN classifiers to perform a hierarchical classification of samples, obtaining comparable accuracies (81 per cent) to the direct 3-class CNN, but considerably worse accuracies in the 4-way case (65 per cent). As an additional verification, we apply our networks to a small sample of Galaxy Zoo images, obtaining accuracies of 92, 82, and 77 per cent for the binary, 3-way, and 4-way classifications, respectively.
Funder
University of Western Australia
Publisher
Oxford University Press (OUP)
Subject
Space and Planetary Science,Astronomy and Astrophysics
Cited by
20 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献