Unsupervised Galaxy Morphological Visual Representation with Deep Contrastive Learning

Galaxy morphology reflects structural properties that contribute to the understanding of the formation and evolution of galaxies. Deep convolutional networks have proven to be very successful in learning hidden features that allow for unprecedented performance in the morphological classification of galaxies. Such networks mostly follow the supervised learning paradigm, which requires sufficient labeled data for training. However, the labeling of a million galaxies is an expensive and complicated process, particularly for forthcoming survey projects. In this paper, we present an approach, based on contrastive learning, with aim of learning galaxy morphological visual representation using only unlabeled data. Considering the properties of low semantic information and contour dominated of galaxy images, the feature extraction layer of the proposed method incorporates vision transformers and a convolutional network to provide rich semantic representation via the fusion of multi-hierarchy features. We train and test our method on three classifications of data sets from Galaxy Zoo 2 and SDSS-DR17, and four classifications from Galaxy Zoo DECaLS. The testing accuracy achieves 94.7%, 96.5% and 89.9%, respectively. The experiment of cross validation demonstrates our model possesses transfer and generalization ability when applied to new data sets. The code that reveals our proposed method and pretrained models are publicly available and can be easily adapted to new surveys. (6)