Classification of low resolution astronomical images using convolutional neural networks

Abstract

Currently deep machine learningtechniques are widely adopted by computer vision and signal processing communities. Deep learning, in particular, the Convolutional Neural Networks (CNN), are the most impressive classifiers widely used for image classification in recent years. CNN model allows the machine to learn automatically about the complex image features from its representation, minimizing the need of human experts in feature extraction. Such a hierarchical representation learning of the images makes CNN a more promising model for classification of different kinds of images as compared to the traditional machine learning models. In this paper, one such successful implementation of CNN is performed for classifying low resolution radio astronomical images containing objects like ‘Radio Halos and Relics’, and several other ‘Point Radio Sources’. For such images, low resolution makes feature extraction a difficult task. Hence, a CNN based classification model proved more efficient in this casegiving a classification accuracy of 88%.