Robust Multi-Class Weather Classification from Images Using Deep Convolutional Neural networks (CNN)

Abstract

This paper presents a robust Convolutional Neural Network (CNN) model designed to classify weather conditions from images into five distinct categories: clear, foggy, rainy, cloudy, and snowy. The model was trained on a well-curated dataset comprising 2,500 images, with an equal distribution across the five categories. The images were resized to 100×100 pixels to standardize input size and optimize training time. The final model achieved an overall accuracy of 85.2%, demonstrating its ability to classify weather conditions effectively. In addition to accuracy, precision, recall, and F1-score were evaluated for each class, showing strong performance across all weather categories. The paper explores the model architecture, training process, evaluation metrics, and provides a comprehensive analysis of the challenges encountered during model development. Finally, the study suggests future directions for improving automated weather classification systems, including the exploration of advanced CNN architectures, the integration of temporal data, and the use of transfer learning.