AI Waste Sorting Assistant

Inspiration In real life, waste sorting introduces cognitive burden for users. Although people understand the rules, they still need time to determine the correct category when facing real objects. Therefore, we aim to build a more intuitive system that reduces this decision-making cost. What it does EcoSort supports three input modes: image upload, camera capture, and text input. The system returns waste classification results, confidence scores, and recycling suggestions to make classification easier. How we built it The backend is built with Python and Flask for handling requests and model inference. The model uses TensorFlow with MobileNetV2 for transfer learning, along with a custom CNN as a baseline model. The frontend is built using HTML, CSS, and JavaScript to support image upload, camera input, and text-based interaction. Challenges we ran into We faced challenges in maintaining consistency across multiple input modes, ensuring stable real-time camera processing, and improving generalization on a small dataset. Accomplishments that we're proud of We successfully implemented a MobileNetV2 classification model, CNN vs MobileNetV2 comparison, real-time camera classification, and a unified multimodal interface. What we learned We learned how transfer learning works in practice, how to integrate frontend and backend systems, and how to design a complete AI interaction pipeline. What's next for EcoSort We aim to improve semantic understanding in text input, enhance real-world robustness, and expand the dataset with more waste categories.

Built With

• cnn
• css
• flask
• html
• keras
• mobilenetv2
• python
• tensorflow