An original approach to automatic classification of auroras by machine identification of images received from sky photo recorders, also known as all-sky imagers, is proposed. The total of 163899 sky images taken at 10-minute intervals within the auroral oval (Kola Peninsula, Russia) were selected over a 10-year period. We propose an intelligent information system designed to classify each acquired image into one of seven predefined categories. Analysis of the quality metrics of the system built on the basis on the ResNet50 neural network architecture showed the accuracy of the classification at the level of 96 %, which is practically unachievable when manually processing data samples of such a volume. The result of automatic classification of sky images based on the proposed system is available at the link: (https://disk.yandex.ru/i/760MyWR4YyVYuw).
A.V. Vorobev (Wed,) studied this question.