Trash2Taste

Inspiration The motivation for this project stemmed from a desire to combat food waste in educational institutions, which is a significant yet often overlooked problem. By integrating smart technology into everyday school operations, we aim to create a sustainable and efficient approach to managing cafeteria waste and optimizing menu planning. What I Learned Throughout this project, I gained deeper insights into the practical applications of machine learning in real-world scenarios. I also improved my skills in data handling, machine learning model tuning, and developing user-friendly interfaces for complex systems. How We Built It Part 1: Smart Menu Generation We began by collecting historical data on meal consumption and associated waste from several school cafeterias. Using this data, we developed a machine learning model that predicts the level of waste for different meals. We then incorporated nutritional guidelines to ensure that all meals meet specific health standards. The final system uses a multi-objective optimization algorithm that balances waste reduction with nutritional fulfillment. Key Technologies: Machine Learning Models (Regression Analysis) Optimization Algorithms Data Analysis Tools Part 2: Smart Trash Can The Smart Trash Can system uses YOLO (You Only Look Once) for real-time food waste classification. We installed cameras and sensors in trash cans to automatically capture and categorize waste. The data is then stored and analyzed to provide actionable insights, which feed back into the menu generation system, completing a data-driven feedback loop. Key Technologies: YOLO for Real-Time Object Detection Embedded Systems (Cameras and Sensors) Data Management and Analysis Solutions Challenges Faced Lack of Publicly Available Datasets: Initially, the absence of specific datasets for food waste in school cafeterias was a major hurdle. We addressed this by creating a synthetic dataset that mimicked real-world waste patterns, allowing us to train our models effectively. Custom YOLO Training Time: The initial training time for our custom YOLO model was around 27 hours, which was impractical for iterative testing and deployment. We managed to reduce the training time to approximately 1 hour by optimizing the model's architecture—reducing the input size, decreasing the number of epochs, and increasing the batch size. Conclusion The combination of the Smart Trash Can and Smart Menu Generation systems introduces a robust solution to food waste and nutritional planning in school cafeterias. By leveraging advanced technologies and innovative data-driven approaches, we not only enhance operational efficiency but also promote environmental sustainability and educational opportunities for students.

Built With

• api
• python