This paper introduces ProtoPNet, a deep neural network architecture for interpretable image recognition. The model reasons by identifying prototypical parts of images and using them to make classifications, similar to how humans explain image recognition. ProtoPNet uses only image-level labels for training without requiring part annotations. It is evaluated on the CUB-200-2011 and Stanford Cars datasets, achieving performance comparable to non-interpretable models and outperforming them when multiple networks are combined. The model provides interpretable reasoning by highlighting parts of the image that are similar to learned prototypes, offering explanations for its classifications. Unlike other interpretable models, ProtoPNet's reasoning process is integrated into the network's architecture, allowing for transparent decision-making. The model is trained using a specialized algorithm that optimizes both the convolutional layers and prototypes, ensuring that prototypes represent meaningful image parts. The paper also compares ProtoPNet with other interpretable models, showing that it provides a unique form of interpretability by linking classifications to similar prototypical parts. The model's reasoning process is visualized and validated through case studies on bird and car classification tasks. The results demonstrate that ProtoPNet achieves high accuracy while maintaining interpretability, making it a valuable tool for deep learning applications where explainability is important.This paper introduces ProtoPNet, a deep neural network architecture for interpretable image recognition. The model reasons by identifying prototypical parts of images and using them to make classifications, similar to how humans explain image recognition. ProtoPNet uses only image-level labels for training without requiring part annotations. It is evaluated on the CUB-200-2011 and Stanford Cars datasets, achieving performance comparable to non-interpretable models and outperforming them when multiple networks are combined. The model provides interpretable reasoning by highlighting parts of the image that are similar to learned prototypes, offering explanations for its classifications. Unlike other interpretable models, ProtoPNet's reasoning process is integrated into the network's architecture, allowing for transparent decision-making. The model is trained using a specialized algorithm that optimizes both the convolutional layers and prototypes, ensuring that prototypes represent meaningful image parts. The paper also compares ProtoPNet with other interpretable models, showing that it provides a unique form of interpretability by linking classifications to similar prototypical parts. The model's reasoning process is visualized and validated through case studies on bird and car classification tasks. The results demonstrate that ProtoPNet achieves high accuracy while maintaining interpretability, making it a valuable tool for deep learning applications where explainability is important.