Morphological Classification of Radio Galaxies using Semi-Supervised Group Equivariant CNNs

Abstract

Out of the estimated few trillion galaxies, only around a million have been detected through radio frequencies, and only a tiny fraction, approximately a thousand, have been manually classified. We have addressed this disparity between labelled and unlabeled images of radio galaxies by employing a semi-supervised learning approach to classify them into the known FRI and FRII types. A Group Equivariant Convolutional Neural Network was used as an encoder that preserves the equivariance for the Euclidean Group E(2) to learn the representation of globally oriented feature maps through new SelfSupervised Learning (SSL) techniques SimCLR and BYOL. After representation learning, we trained a fully-connected classifier and fine-tuned the trained encoder with labelled data. We have found that this semi-supervised approach helps our method outperform a state-of-the-art method of classifying radio galaxies in many metrics. Our work reiterates the importance of semisupervised learning in radio galaxy classification, where labelled data are scarce, but prospects are immense.