Images Based Classification for Warm Cloud Rainmaking using Convolutional Neural Networks

Abstract

This paper is an experiment conducted on deep learning technique, which is currently popular and efficient in the field of image classification. The Convolutional Neural Networks (CNNs) has been achieved in many tasks of image classification and used to classify cloud images for meteorology task. The CNNs can be applied to classify rainmaking cloud images. There are three steps of warm clouds: fattening, attacking, and enhancing. The challenge in this work is the images required for classification are very similar. That means the sampling images are the images of cloud in the same type but difference at the formation stage, while the image shooting format and equipment is not fixed. The images are taken with varying degrees of angle, height, lightness and resolution. From the experimental results, it was found that CNNs had prominent characteristics for learning to extract the necessary attributes used in solving problem along with self-classification, leading to the developed model had higher accurate classification than traditional method around 5–8%.