Rotation-invariant convolutional neural networks for galaxy morphology prediction

Abstract

Measuring the morphological parameters of galaxies is a key requirement for studying their formation and evolution. Surveys such as the Sloan Digital Sky Survey (SDSS) have resulted in the availability of very large collections of images, which have permitted population-wide analyses of galaxy morphology. Morphological analysis has traditionally been carried out mostly via visual inspection by trained experts, which is time-consuming and does not scale to large ( 10 4 ) numbers of images.Although attempts have been made to build automated classification systems, these have not been able to achieve the desired level of accuracy. The Galaxy Zoo project successfully applied a crowdsourcing strategy, inviting online users to classify images by answering a series of questions. Unfortunately, even this approach does not scale well enough to keep up with the increasing availability of galaxy images.We present a deep neural network model for galaxy morphology classification which exploits translational and rotational symmetry. It was developed in the context of the Galaxy Challenge, an international competition to build the best model for morphology classification based on annotated images from the Galaxy Zoo project.For images with high agreement among the Galaxy Zoo participants, our model is able to reproduce their consensus with near-perfect accuracy (> 99%) for most questions. Confident model predictions are highly accurate, which makes the model suitable for filtering large collections of images and forwarding challenging images to experts for manual annotation. This approach greatly reduces the experts' workload without affecting accuracy. The application of these algorithms to larger sets of training data will be critical for analysing results from future surveys such as the LSST.

Deep learning for studies of galaxy morphology

“…Such accuracy has been obtained both in the work from Dieleman, Willett, & Dambre (2015) using data from the Galaxy Zoo project and at high redshift in Huertas-Company et al (2015) for CANDELS fields (Galametz et al 2013) galaxies. Two-dimensional profile fitting is often used to measure the properties of galaxies .…”

ASTErIsM: application of topometric clustering algorithms in automatic galaxy detection and classification

“…Banerji et al (2010) and Dieleman et al (2015) have used artificial neural networks (ANN), while HuertasCompany et al (2008);Huertas-Company et al (2011) have used Support Vector Machine (SVM) classifier and Ferrari et al (2015) have used Linear Discriminant Analysis (LDA) classifier. Banerji et al (2010) have used a sample of 75000 object (50000 for training and 25000 validation) from the Galaxy Zoo 1 catalog, to train a neural network in order to reproduce the human classification in early type, spiral, and point source/artifact .…”

“…Performance of Banerji et al (2010) are comparable with our results, with the difference that we use a larger amount of features, but we do not use color information. Dieleman et al (2015) used a a training set of 61578 images, and 79975 images for the validation, from Galaxy Zoo 2, to train their neural network to reproduce the human answers to the 11 tasks in Galaxy Zoo 2. The results presented by the authors refers only to a sub sample of the test set images with at least 50 percent of participant answering.…”

“…The results presented by the authors refers only to a sub sample of the test set images with at least 50 percent of participant answering. In our work we have investigated two task in common with Dieleman et al (2015): t01 and t04. Dieleman et al (2015) evaluated their classification metrics in 10 different bins of participants agreement.…”

“…neural networks (Naim et al 1995;Lahav et al 1996;Goderya & Lolling 2002;Banerji et al 2010;Dieleman et al 2015) , decision trees (Owens et al 1996) and ensembles of classifiers (Bazell & Aha 2001).…”