Speaker
Description
This paper presents the development of a real-time weather classification system using a hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) architecture. The system integrates visual data from surveillance cameras and time-series weather sensor data to classify weather conditions into sunny, cloudy, and rainy categories. Two data alignment strategies were explored to investigate the impact of label assignment methods on classification performance. The first approach assigned labels to images based on visual interpretation, and then matched them with sensor data. The second approach used sensor-derived labels and matched them to the corresponding image frames. Experimental results demonstrated that the second approach offered better generalization and real-world robustness, achieving an overall accuracy of 90 percent. Evaluation metrics such as confusion matrix, accuracy, MAE, and RMSE were used to validate the system's performance. The proposed model is suitable for short-term weather classification tasks and has potential for deployment in real-time weather monitoring systems.
