Pokémon Classification Using Convolutional Neural Networks: A Deep Learning Approach for Fine-Grained Image Recognition

The project below classifies Pokémon species using Convolutional Neural Networks, which have high inter-class similarities and complex visual features. We employed a dataset from Kaggle containing over 7,000 labelled Pokémon images and then designed and optimized a CNN model for multi-class classification. Preprocessing on the dataset included image resizing, augmentation, and normalization to improve the robustness of the model. Our model architecture consists of several convolutional layers with max-pooling, followed by a fully connected layer, optimized using the Adam optimizer and categorical cross-entropy loss function. The model achieved an impressive classification accuracy of 95.8%, showing its capability to distinguish between 150 different Pokémon species effectively. Evaluation metrics, including precision, recall, and F1-score, further validated its high performance. The key takeaways from this work indicate that CNNs can perform well on fine-grained image classification and form a foothold for the next research endeavors on deep learning-based visual recognition challenges.