Automatic Cloud Classification Using Neural Networks with SVMS in Ground-Based Sky Images

Forecasting the generation of renewable energy and investigating climate change require the prediction and classification of cloud cover (CC). In enhancing the accuracy of various weather predictions and the efficiency of renewable energy generation, cloud recognition from low-level aerial images is critical. This research's integrated approach uses CNNs and SVMs to classify cloud cover from sky photos captured by all-sky cameras. The CNN-SVM hybrid model is proposed to increase accuracy in cloud classification through 49,016 human-classified pictures obtained from various regions in Spain. Cloud classes are properly segregated using the SVM classifier while the CNN component extracts the high-level features. Experimental results demonstrate that the proposed model outperforms other models based only on CNN architectures, with over 98% accuracy in classifying various types of clouds. As new data from different geographic zones, including an Antarctic station, have been added to the model, the validity of the same is confirmed to test the general validity under different climatic scenarios. Getting an average accuracy of 98.09%, the paper also explores the combination of EfficientNet and ResNet to promote feature extraction and classification. For the innovative and enhanced methodology of systemic cloud categorization, deep learning enhances traditional forms of machine learning, including SVMs. This framework is crucial for climate change as well as renewable energy management.