Kyushu University
National Institute of Polar Research, Research Organization of Information and Systems (NiPR)
National Institute of Polar Research, Research Organization of Information and Systems (NiPR)
Kyushu University
出版者
宇宙航空研究開発機構(JAXA)
出版者（英）
Japan Aerospace Exploration Agency (JAXA)
雑誌名
宇宙航空研究開発機構研究開発報告: 宇宙科学情報解析論文誌: 第4号
雑誌名(英)
JAXA Research and Development Report: Journal of Space Science Informatics Japan: Volume 4
Aurora, which is attractive for many people, is an astronomical phenomenon related to many fields, such as interplanetary space, magnetosphere and ionosphere, and thus it is difficult to predict behaviors or shapes of aurora. The big goal of this research is to forecast them using observed data in different formats from different fields. To do that, we need training data which shows when and what types of aurora have appeared. In this paper, we evaluate three popular methods of automatic classification of images to classify auroral all-sky images. We found that two methods based on the local feature and the color histogram, both of which are expected to classify auroral all-sky images in detail, fail to capture characteristics of aurora in the preliminary experiment, while we obtained 92.3% of the classification accuracy based on auroral area, which can only classify whether an image contains aurora or not.
内容記述
形態: カラー図版あり
内容記述（英）
Physical characteristics: Original contains color illustrations
オーロラの出現・形状の予測に向けた全天オーロラ画像の自動分類への試み
AA1530026010.pdf
(1.56MB)
