Automatic classification of mesoscale auroral forms using convolutional neural networks. (1st September 2022)

• Record Type: Journal Article
• Authors: Guo, Z.-X.; Yang, J.-Y.; Dunlop, M.W.; Cao, J.-B.; Li, L.-Y.; Ma, Y.-D.; Ji, K.-F.; Xiong, C.; Li, J.; Ding, W.-T.
• Is Part Of: Journal of atmospheric and solar-terrestrial physics. Volume 235(2022)
• Publication Date: 2022-09-01
• Subjects: Geophysics -- Periodicals; Atmospheric physics -- Periodicals; Géophysique -- Périodiques; Météorologie physique -- Périodiques; Electronic journals; 551.51
• DOI: 10.1016/j.jastp.2022.105906
• Languages: English
• ISSNs: 1364-6826

Abstract

Abstract: Convolutional neural networks (CNNs) in deep learning enable the extraction of features in image data. Through the multi-layer superposition of a convolutional neural network, we can better capture the essential characteristics of different auroral subclasses and further classify auroral images in detail. Because the auroral morphological features often present abstract characteristics, our study compares different CNN architectures and different layering in order to test the best neural network model for mesoscale aurora classification. Although the classification models and subclasses used by us are both more complex, the highest F1 score of aurora classification of the test set reaches 99.6% (ResNet-50), which performs best comparing with previous works. Our classification models work also quite well when applied to an independent auroral image sequence, declaring our approach can automatically select images of various mesoscale auroral forms using CNNs, and allow the time sequence of auroral evolution to be seen automatically through the mesoscale auroral feature recognitions. Highlights: The time sequence of auroral evolution is shown automatically by the mesoscale auroral feature recognitions for the first time. Abstract : Convolutional neural networks (CNNs) in deep learning enable the extraction of features in image data. Our study compares different CNN models in order to test the best neural network model for mesoscale aurora classification. Although the classification models and subclasses used by us are both more complex, the highest F1 score of aurora classification of the test set reaches 99.6% (ResNet-50), which performs best comparing with previous work. Our classification models work also quite well when applied to an independent auroral image sequence, declaring our approach can automatically select images of various mesoscale auroral forms using CNNs, and allow the time sequence of auroral evolution to be seen automatically through the mesoscale auroral feature recognitions.