Sort-omatic

Sort-omatic is a smart trashcan that automatically sorts garbage while also having a social impact by incentivizing recycling and composting through leaderboards, statistics, and encouraging messages.

Comment

What it is

Sort-omatic is a smart trash can that automatically sorts garbage into trash, recycle, and compost - while also having a social impact by incentivizing recycling and composting through tools such as leaderboards, statistics, and encouraging messages.

What it does

Using AI and Machine Learning, this device can detect various types of trash and classify them into trash, compost, and recycle - so that less waste ends up in landfills, and more waste ends up being beneficial to society. This device also has a fully-functional website that serves as an incentivizing tool - it displays statistics about trash sorting based on each account, with a points-system and leaderboard: so that more people are encouraged to sort trash.

How we built it

The website was built using technologies such as Node.JS, Express, HTML, CSS, JS, EJS, Chart.js, and SVG.

For our model, we developed a convolutional neural network from scratch which used a combination of convolution, max pooling, flatten, and dense layers with rectified linear and SoftMax activation functions, all provided by the TensorFlow Keras library developed by Google.

We used various Google Cloud Products such as Firebase, Cloud Compute Engine, Cloud TPU API, and Cloud Storage to create our AI Convolutional Neural Network and Website.

Challenges we ran into

  • rescaling SVGs was very hard to do as there was very little documentation on how to do so.
  • getting real-time data from Firebase, although we were able to figure it out!
  • creating our first-ever "full" website with functional log-in , register and account features (although this is also something we are proud of :D)
  • Using multiple libraries where a lot of them did not have any documentation
  • Libraries
  • CADding the Sort-omatic designs took a very long time because of the lack of tools available
  • Connecting our code to Google Cloud's Linux VM's to train our model.
  • Uploading our training dataset to Cloud Storage Buckets.

Accomplishments that we're proud of

  • Being able to complete both the website and AI Model on time!
  • Using Firebase and all the other Cloud tools for the first time!
  • Creating our first-ever "full" website with functional log-in , register and account features!
  • Successfully CADding for the first time!
  • Deploying a model with 95+% testing accuracy.
  • Using python libraries such as cv2 to do real time image classification

What we learned

  • How to use Firebase and other Google Cloud Products/API's
  • Creating our first full website
  • Integrating Machine Learning and both front-end and back-end Web Development together to create a full stack application

What's next for Sort-omatic

  • Expanding the social aspect of our project, so that people invite their friends to Sort-omatic and we become more popular
  • Training on more datasets with more images to further increase the accuracy of our model
  • Create an actual protype of Sort-omatic using Raspberry Pi.
Share this project:

Updates