Radio Galaxy Zoo: compact and extended radio source classification with deep learning

Abstract

Machine learning techniques have been increasingly useful in astronomical\napplications over the last few years, for example in the morphological\nclassification of galaxies. Convolutional neural networks have proven to be\nhighly effective in classifying objects in image data. The current work aims to\nestablish when multiple components are present, in the astronomical context of\nsynthesis imaging observations of radio sources. To this effect, we design a\nconvolutional neural network to differentiate between different morphology\nclasses using sources from the Radio Galaxy Zoo (RGZ) citizen science project.\nIn this first step, we focus on exploring the factors that affect the\nperformance of such neural networks, such as the amount of training data,\nnumber and nature of layers and the hyperparameters. We begin with a simple\nexperiment in which we only differentiate between two extreme morphologies,\nusing compact and multiple component extended sources. We found that a three\nconvolutional layer architecture yielded very good results, achieving a\nclassification accuracy of 97.4% on a test data set. The same architecture was\nthen tested on a four-class problem where we let the network classify sources\ninto compact and three classes of extended sources, achieving a test achieving\na test accuracy of 93.5%. The best-performing convolutional neural network\nsetup has been verified against RGZ Data Release 1 where a final test accuracy\nof 94.8% was obtained, using both original and augmented images. The use of\nsigma clipping does not offer a significant benefit overall, except in cases\nwith a small number of training images.\n