Auroral Image Classification With Deep Neural Networks

Abstract

<p>Results from a study of automatic aurora classification using machine learning techniques<br />are presented. The aurora is the manifestation of physical phenomena in the ionosphere-magnetosphere<br />environment. Automatic classification of millions of auroral images from the Arctic and Antarctic is<br />therefore an attractive tool for developing auroral statistics and for supporting scientists to study auroral<br />images in an objective, organized, and repeatable manner. Although previous studies have presented<br />tools for detecting aurora, there has been a lack of tools for classifying aurora into subclasses with a<br />high precision ( > 90%). This work considers seven auroral subclasses: breakup, colored, arcs, discrete,<br />patchy, edge, and faint. Six different deep neural network architectures have been tested along with the<br />well-known classification algorithms: k-nearest neighbor (KNN) and a support vector machine (SVM).<br />A set of clean nighttime color auroral images, without clearly ambiguous auroral forms, moonlight,<br />twilight, clouds, and so forth, were used for training and testing the classifiers. The deep neural networks<br />generally outperformed the KNN and SVM methods, and the ResNet-50 architecture achieved the highest<br />performance with an average classification precision of 92%.<br /></p>