Terra

AI/Computer Vision Waste Detection App

Comment 1

Inspiration

Our team wanted to build an environmentally-impactful application. Our research naturally led us to recognize trash as being one of the most significant contributors to a range of environmental issues; climate change, biodiversity loss, and chemical pollution.

Our goal was to create an app that both drove users to be more conscious and aware of their waste reduction methods and have the possibility of streamlining massive waste reduction strategies in the future.

What it does

Terra is an app that allows users to take a picture for a machine learning model to predict the possibility of the item pictured being trash, recyclable, or compostable. In addition to this, the app contains local resources for users to recycle and compost more efficiently.

We hope that both of our goals were achieved through this.
1) Since 62% of Americans are unable to properly recycle. This model is trained on common mistakes, mistakes such as recycling used paper/aluminum/certain snack packaging, in order to teach users proper how to properly recycle and compost and holistically bring more awareness of the need to employ waste reduction strategies.
2) The ~90% precision rate of the model is an excellent tool for greater projects in streamlining everyday waste. For the 96% of waste that is not properly recycled, a machine learning model is an important tool in the future of waste reduction strategies.

How we built it

• We used Google's new Vertex API, on the Cloud, in order to train our model. The dataset is linked in the Github.
• React Native was used on the front-end in order to build the app.
• Google cloud expo packages were used for the local waste reduction resources.

Challenges we ran into

The iPhone camera sends images that are extremely large and too high quality. The hardest part of this project was restricting our tools to that of React Native Expo packages in order to make the proper API requests. We almost considered building an entire Django backend in order to both process the image and process the API request. The mentors at KnightHacks were extremely helpful in allowing us to leverage expo React Native to the max.

Accomplishments that we're proud of

Our team has not been exposed to a machine learning model before and thanks to Google Cloud's Vertex API, we touched something we never thought we would ever venture into.

Additionally, our team has never touched React Native before. We originally wanted to settle with Swift and focus on IOS development, but technical difficulties forced us to learn React Native and build our entire app around it.

What we learned

We learned a lot about mobile app development and the moving parts in a machine learning model.

What's next for Terra

We hope to make the app more android friendly and to also expand the resources for waste reduction. A feature we would've liked to implement is crowdsourced local composting stations for the community as a whole to contribute their empty composts for the community to donate trash too.

Share this project:

Updates