A Hybrid Approach for Cloud Image Classification Based on Data Augmentation and Transfer Learning

The prediction of rainfall is greatly affected by the condition and composition of clouds in the atmosphere. Clouds are generally composed of ice crystals, water grains, or both, and have a direct influence on climate, ecological equilibrium, air travel etc. As a result, the classification of clouds has become an important and captivating subject in meteorology and garnering significant attention from researchers. This paper presents a robust methodology in cloud image classification by leveraging transfer learning and data augmentation. Data augmentation is employed to equalise the dataset, generating additional instances by horizontally flipping, distorting, introducing noise, or extracting patches from the original images. A fine-tuned VGG16 model with a K-fold cross-validation technique is utilized for high-level feature extraction. The experiments are performed on the Singapore Whole Sky Imaging Categories (SWIMCAT) dataset. Also, the different evaluation metrics are used to evaluate the usefulness of the model. It is found that the accuracy of the balanced datasets has been improved by 0.53%.