구름 분류 방법에 대한 연구

Abstract

This study is based on the mentorship course in CNU Science Education Institute for the Gifted. Clouds are classified into ten types proposed by Luke Howard and documented in the Cloud Atlas. In this study, we performed the cloud classification using manual approach and machine learning techniques. A total of 365 cloud images were directly taken for the mentorship course. Initially, we visually assessed cloud features in each image and manually classified cloud images into 11 types considering cloud features. The 11 types were named according to their distinguishing features. Subsequently, cloud features in each image were quantified based on the HSV color space. We applied the quantified cloud features to decision tree and k-means clustering algorithm, resulting in classification of clouds into six and three types, respectively. The disparity in the number of classifications between manual and machine learning methods is originated from the relative less features derived from the cloud images and discrepancies between human intuitive criteria and those used in machine learning classification. In the further study, by utilizing a more extensive set of cloud features and samples, well-trained machine learning techniques will allow for a more precise classification of clouds. This is expected to enable the utilization of these results in short-term weather forecasts.