Weather Image Classification Using VGG19

The VGG19 deep learning model is utilized in this study to evaluate weather classification; transfer learning is conducted to increase performance. The collection comprises roughly 900 weather photographs across five categories: overcast, foggy, rainy, shine, and dawn. The VGG19 model was updated to identify unique weather patterns after being pre-trained on ImageNet. Key hyperparameters include a sample size of 32, a learning rate of 1e-4, and the Adam optimizer, which was selected for its efficacy in gradient-based optimization. The model's end loss on the test set was 0.28 after it was trained for a number of epochs to minimize the categorical cross-entropy loss. After training, the model gained an accuracy of 92.67%, showing its capacity to generalize successfully on unknown input. This method works better than many traditional CNN algorithms and provides a workable solution for weather classification problems.