Replication Data for: Auroral Image Classification with Deep Neural Networks

Results from a study of automatic aurora classification using machine learning techniques are presented. The aurora is the manifestation of physical phenomena in the ionosphere magnetosphere environment. Automatic classification of millions of auroral images from the Arctic and Antarctic is therefore an attractive tool for developing auroral statistics and for supporting scientists to study auroral images in an objective, organized and repeatable manner. Although previous studies have presented tools for detecting aurora, there has been a lack of tools for classifying aurora into subclasses with a high precision (>90%). This work considers seven auroral subclasses; breakup, colored, arcs-bands, discrete, patchy, edge and clear-faint. Five different deep neural network architectures have been tested along with the well known classification methods; k nearest neighbor (KNN) and support vector machine (SVM). A set of clean nighttime color auroral images, without ambiguous auroral forms, moonlight, twilight, clouds etc., were used for training and testing. The deep neural networks generally outperformed the KNN and SVM methods, and the ResNet-50 architecture achieved the highest performance with an average classification precision of 92%. Although the results indicate that high precision aurora classification is an attainable objective using deep neural networks, it is stressed that a common consensus of the auroral morphology and the criteria for each class needs.<br><br>The authors would like to thank Urban Brändström and the Swedish Institute of Space Physics for providing the original auroral image data. The image data archive is freely accessible at http://www2.irf.se/allsky/data.html, however, the users are obliged to contact the Kiruna Atmospheric and Geophysical Observatory before usage

本研究展示了一项基于机器学习技术开展自动极光（Aurora）分类的研究成果。极光是电离层-磁层环境中物理过程的表现形式。对来自北极与南极的数百万幅极光图像进行自动分类，可为构建极光统计体系提供有力支撑，同时帮助科学家以客观、规范且可复现的方式开展极光图像研究。尽管既往研究已推出极光检测工具，但仍缺乏可将极光划分为子类且分类精度大于90%的专用分类工具。本研究涵盖七类极光子类：爆发型（breakup）、彩色型（colored）、弧带型（arcs-bands）、离散型（discrete）、斑块型（patchy）、边缘型（edge）以及清晰弱型（clear-faint）。本研究测试了五种不同的深度神经网络（Deep Neural Network, DNN）架构，同时搭配经典分类方法：k近邻（KNN）与支持向量机（SVM）。本研究采用一组无模糊极光形态、月光、暮光、云层等干扰因素的高质量夜间彩色极光图像，用于模型的训练与测试。整体而言，深度神经网络的分类性能优于KNN与SVM方法，其中ResNet-50架构表现最优，平均分类精度达92%。尽管研究结果表明，借助深度神经网络可实现高精度极光分类，但需强调的是，目前学界尚未就极光形态学分类标准及各类别的划分依据达成统一共识。

作者谨此感谢Urban Brändström及瑞典空间物理研究所（Swedish Institute of Space Physics）提供原始极光图像数据。该图像数据集存档可于http://www2.irf.se/allsky/data.html 免费获取，但使用者需在使用前联系基律纳大气与地球物理观测站（Kiruna Atmospheric and Geophysical Observatory）。