SmartRecycle Kiosk

Classified plastic material
Classified metal material
Produced recyclable fun fact
3D Model of proof of concept kiosk

Inspiration

College campuses are continuing to integrate recycling programs on their campuses. However, the multiple recycling bins and vague descriptions cause confusion among users leading to materials being misplaced or contaminated. This reduces the efficiency of recycling programs and increases burden on waste management systems.

What it does

Our project aims to address this issue by simplifying the recycling process, providing a more efficient and faster interaction for users. By analyzing images of objects, our project can determine whether an item is paper, plastic, metal, or another recyclable material, and provide clear guidance on which recycling bin it should go into.

How we built it

The frontend and backend was developed using React/Node. ResNet18 neural network model was pre-trained on ImageNet dataset and was fine tuned over recyclables dataset. The model was deployed via Python FastAPI service. Recyclable fun facts were derived from Gemeni API calls incorporating model classification in prompt.

Challenges we ran into

Teaching our machine learning model to correctly identify a wide range of items was difficult due to unusual shapes, labels, or mixed materials. In order to accurately identify the material of objects, we had to find datasets to train our model on. However, the datasets that were available were unlabeled or labeled using unrealistic images. Variations in lighting, angles, and camera resolution affect the accuracy of image recognition.

Accomplishments that we're proud of

We are proud of the smooth integration of subsystems in our project. We are proud of our 3D model of proof of concept kiosk. We are proud of the accuracy of our project in classifying the material of objects.

What we learned

We learned how to deploy python scripts inside a node.js/React web application.

What's next for SmartRecycle Kiosk

Improve design: Actual kiosk implementation, accessible for various sizes and materials Statistics counter: Allow users to view usage of materials, Allow larger establishments to plan better for events by reducing waste Mobile Use: Enable back camera support for better quality and quicker access Beta-testing: Get feedback from users about accuracy and interface usability