This research explores the use of CNNs for image classification tasks, leveraging PyTorch for model development and training. The research emphases on the fundamental layers of CNN detailing their roles in classification. A CNN architecture was designed and trained using PyTorch, optimizing key hyperparameters include various convolution layers, activation functions, kernel size, padding, dropout, batch size, and learning rate. Study demonstrates the performance of CNNs in handling complex image data while addressing challenges like overfitting, dataset imbalance, and computational efficiency. The results show that proper preprocessing and hyperparameter tuning significantly improve classification accuracy, making CNN-based models more adaptable to real-world applications.