Inspiration

A common nuisance in the summer: the spotted lanternfly. We initially set out to identify and quash these pests using an image classification algorithm. However, when presented with the challenge prompts we thought, "How can we use the same image classification algorithm to improve campus life and quash another nuisance?"

The answer was through sorting recyclables. Penn State has taken the initiative to improve its solid waste stream, but still finds high amount of contaminants in bins (Misc Plastics bins contained only 40% target material in 2019 waste stream audit). These contaminations are very costly, with fines of up to $250/ton as opposed to standard $20/ton fee for sorted recyclables. (PSU collects over 2000 tons of recyclables a year)

By taking the burden of identification off of students, we can both improve the trash collection experience and save the Penn State resources that could be redirected to the student life.

What it does

Webapp receives a picture and helps you identity trash. From there a image classification model determines its material and appropriate bin for disposal.

How we built it

- Our webapp is built to be platform agnostic though react-native
- Image classification algorithm is resnet34 fastai modulated in Colab
- Deployed using Render

Challenges we ran into

Surprisingly, it was the deployment of our AI model that was the most troublesome part. We ran into memory limitations with Heroku and in Azure we had great difficulty in attempting to host non-native AI models. Finally we opted to use Render for rapid deployment.

Accomplishments that we're proud of

 - Coming in as individuals and building a cohesive team.
 - Creating a full stack application in under 24 hours
 - Learning new platforms

What we learned

 - To collaborate as a team.
 - Create a workflow for rapid application development
 - New platforms: Azure, react-native, MS Power Apps

What's next for Waste Sorting Using Computer Vision

Next step is to integrate and collaborate with other ecofriendly products such as LOMI (compost). There is also incentive for automation, turning the 5+ bins at every location to 1 bin that sorts automatically with using embedded software.