This project presents a food classification system using a Deep Separable Convolutional Neural Network (DS-CNN) architecture based on ShuffleNet. By decomposing standard convolutions into depthwise and pointwise operations, the model significantly reduces computational cost while maintaining high accuracy. Leveraging ShuffleNet’s channel mixing capabilities enhances information flow and representation. The model achieved 88.3% accuracy on a custom food dataset and was evaluated using metrics like precision, recall, and F1 score. A user-friendly GUI was also developed to allow intuitive interaction with the model for real-time classification. This system offers potential applications in nutritional analysis and dietary monitoring.
The DS-CNN model, powered by ShuffleNet, demonstrates efficiency and robustness for food classification tasks. It offers substantial reduction in model size and computational requirements while delivering competitive accuracy. The intuitive GUI further enhances accessibility and potential integration into dietary planning systems.
Model performance may be affected when applied to more diverse or real-world food image datasets. The system also currently relies on pre-labeled categories and may require adaptation for broader food classification use cases.
Future improvements will focus on dataset expansion, multi-label classification, cloud deployment for mobile access, and incorporation of nutritional databases. The project also plans to explore real-time calorie estimation using food image metadata.