GreenAI

Google Slides Link

Inspiration

Climate change has been around for about 800,000 years. Manufacturing plants for making products are the main cause of this problem. One of the main solutions to this problem is recycling. This is well known in our schools and communities, however, people lack the incentive to help fight against it. People know what to do to help the community but don’t do it. A solution needs to be found to encourage people to make our community a better place.

What it does

My solution is a game where people can log their recycling progress and compete with each other. They do this by taking a photo each time they recycle and uploading it to the game. Artificial intelligence is used to process the image and calculate the carbon saved. People can then view the leaderboard to see how they recycle compared to others in their community. This then provides a push for them to start recycling so they can move up on the leaderboard.

How I built it

For the front end, I used HTML, JavaScript, and CSS. Using those languages I was able to create a multipage website that showcases to the user everything they need. For the back end, I used Python and Tensorflow to run inferences on the recyclable items and return how much carbon dioxide was saved. To connect the frontend and backend I used Google Drive as a database.

Social Impact

Just like how video games pull people into playing more with friendly competition, my game creates a competitive desire to recycle more so that users can beat other players’ scores. It channels the human competitive spirit to motivate everyone to recycle. This simultaneously helps the community become a better place while making people feel happy about it. Every person lends a helping hand in protecting our community. This solution is also sustainable because there will constantly be more and more people joining the app as it spreads through the community, meaning more people joining the fight to protect the community.

AI/ML Usage

The artificial intelligence model used is a Mask R-CNN model trained on Google’s CircularNet waste management database. The inference is done with the Tensorflow Object Detection API and a Python script running on a backend server. The model is fed a post-processed picture from the user’s camera and then runs inference on it in ~10 seconds. This model can detect 34 kinds of waste ranging from bottles to cans to paper products. With images of waste at the event, this model detects with an accuracy of ~99%. The user gets a visualisation of the objects the model classified and its accuracy which are intuitive visualisations.

Tech Stack

• The user takes a picture and uploads it to the game
• The game sends the picture to the backend via Google Drive
• The backend runs object detection inference on the picture
• The backend sends a JSON list of the objects as well as a labeled image back to the game
• The game shows the user their score and updates it on the leaderboard
• All data is deleted after the user takes a look
• Frontend → HTML, CSS, and Javascript
• Backend → Python and TensorFlow

Challenges I ran into

The AI detection did not detect all of the bottles or cans in an image. For example, in a group of two bottles, it would only detect one bottle, thus not giving an accurate carbon score. I overcame this by training the model longer and adjusting the post-processing of the image to make the items stand out more from the background.

Accomplishments that I'm proud of

I am proud to make and use an AI model that can accurately detect recyclable items in an image. I am also happy with myself that I was able to make the website polished with so many pages. I was able to make a database connection between the front and backend that works on any device which is something that is not very easy. Lastly, I am fortunate to be able to put something like this out in the world, as I hope that this becomes something that teens and adults enjoy and something we can implement in our daily lives. After all, we have to protect the earth we live in.

What I learned

I learned and acquired numerous skills throughout the making of this project. First I learned how to research a topic that interests me and has a real-life use case and purpose. I also learned how to manage my time quickly as since I was working solo I had to design the frontend, create the AI object detection for the backend, and link them all together, as well as make the presentation. Last but not least, I had so much fun learning how to style the website and design a logo just for it! Picking out theme colors, and picking out a logo added the dash of spice it needed.

What's next for GreenAI

• Increase publicity by spreading it to schools, offices, etc
• Adding more data so that the model can detect more accurately
• Having weekly competitions in the game to spark more interest in climate change and sustainability
• Developing a carbon calculator addon so that people can calculate their score using other factors other than just recycling(gas, food, electricity, etc.)
• Detecting if the image is on a computer screen to prevent cheating

Built With

• css3
• html5
• javascript
• python
• tensorflow