Revolutionizing Meteorological Forecasting with Cloud Image Analysis Using Edge Flow and YOLO

This paper focuses on the development of a Meteorological Service for Air Navigation system, which classifies clouds into five categories: BKN, CAVOK, FEW, OVER, and SCT. The classification is based on the cloud patterns captured in images, which are interpreted as the sky coverage measured in oktas, dividing each image into eight parts. The main contribution of the proposed method lies in the use of the Edge Flow algorithm to convert images into vector data, which are then analyzed and modeled for forecasting using the YOLO algorithm. The images, collected continuously both day and night, are processed to provide weather forecasts, which in turn facilitate weather alerts. The methodology of the study consists of three main steps. The first step involves using the Edge Flow algorithm to transform the images into vector form. The second step focuses on feature extraction and training the model using cloud images categorized into the five defined types. Finally, the third step is the evaluation of the model’s accuracy through the testing of weights generated during training. In this study, over 10,000 images were used for the experiments, yielding average mAP, precision and recall scores of 84.52%, 77.46%, and 86.84%, respectively.