This project focuses on weather image classification task using a convolutional neural network (CNN). Here is a summary of the code:

Dataset Download: Two datasets are downloaded from Kaggle using the opendatasets library.
The first dataset is from https://www.kaggle.com/datasets/jehanbhathena/weather-dataset.
The second dataset is from https://www.kaggle.com/datasets/vijaygiitk/multiclass-weather-dataset.

Data Loading: The first dataset is loaded using tf.keras.preprocessing.image_dataset_from_directory.
The second dataset is used to test the model on new data.

Data Augmentation and Preprocessing: Image data generators are created with data augmentation techniques (e.g., horizontal flip, rotation, zoom) and normalization.
Training and validation datasets are generated using flow_from_directory.

Model Architecture: The model architecture is a combination of a pre-trained ResNet152V2 base model and additional custom layers.
The pre-trained ResNet152V2 model is loaded with weights frozen.
Additional layers, including convolutional layers, LeakyReLU, MaxPooling2D, GlobalAveragePooling2D, Dense, Dropout, and Activation layers, are added to create the complete model.

Model Compilation: The model is compiled with the Adam optimizer and sparse categorical crossentropy loss.

Training: The model is trained on the first dataset using the fit method, with callbacks for model checkpointing and early stopping.

Testing on New Data: The model is tested on new data from the second dataset.
Some sample images from the "rainy" class are loaded and predictions are made.

Evaluation and Visualization: Confusion matrices and class percentages are computed for the "rainy" and "foggy" classes.
Heatmaps are generated to visualize confusion matrices.

Testing on Other Weather Classes: The model is tested on images from the "sunny" class.