Classification of Ground-Based Cloud Images Using EfficientNet-B0: A Study on the CCSN Dataset

Abstract

Clouds cover more than 60% of the Earth's surface and play an important role in the hydrological cycle, climate system, and radiation balance by altering shortwave and longwave radiation. The accuracy of weather forecasts is critical for many sectors, including aviation, maritime transport, agriculture, energy, and environmental monitoring. In this study, a deep learning-based approach was developed using the EfficientNet-B0 architecture for the classification of ground-based cloud images. When using the original Cirrus Cumulus Stratus Nimbus (CCSN) dataset, which contains 2543 images, the model's accuracy rate remained at 53%. However, when the number of images for each cloud class was balanced to 1,000 using data augmentation techniques, a significant increase in model performance was observed, with the accuracy rate reaching 90.14%. The results obtained demonstrate that the EfficientNet-B0 architecture delivers effective performance in cloud classification tasks when data balance is achieved, offering a promising solution for meteorological analysis, aviation, and climate observation applications.