Recyclabot

Main page
Neural network guessing your picture
Incorrect guess; ask for user advice
Global statistic view

Inspiration🤖♻️🤖♻️

We live in an era of convenience. From needing a ride across town to talking to someone from the across the world, a person need only depend on the small devices that they carry on them. Never before could so many things be done with just a simple touch of a button. But how can this technology of convenience be integrated to help with the rising issue of our landfills overflowing? One of the main reasons that the typical American does not recycle is simply because they do not know how to. Without the knowledge of what is and isn't recyclable, we risk potential recyclable goods ending up in landfills, adding to the giant masses of waste and decay or we risk non-recyclable goods contaminating the recyclable ones, which resorts to them being sent to aforementioned land fills! We saw this as an opportunity to use the power of machine learning to help solve this problem. A way to make recycling convenient.

What it does

Recyclabot is a website that can be used to help with sorting through recyclables vs non-recyclables. All a user needs to do is upload an imagine and Recyclabot is able to distinguished whether or not the item in the photo belongs in the recycling bin. Recyclabot will also learn from user interaction so with every use, it will become more and more effective. Recyclabot also keeps track of user interactions and their location, presenting the data of what items are recycled globally.

How we built it 🔨

The whole project can be broken down into two components, the web server interface and the underlying neural network. Firstly the web server component uses flask as its framework, the web server handles all of the user interactions and the file uploading. The neural network component is a custom trained neural network. Currently, its accuracy is quite low at the moment (~60%), it will become more accurate as users submit their images and corrects the algorithm. The neural network is capable of recognizing six classes with four being recyclable types, one compost, and a non-recyclable.

Challenges we ran into

Building the server with HTML was definitely the biggest challenge to our project. In fact the first day was entirely spent on learning the basic of HTML language. But regardless of the headache, the final HTML product is something that we are quite proud of, especially being newbies.

Accomplishments that we're proud of

We are proud of the project as a whole. Despite having only two days to create the entire program, we think it turned out quite well. Especially with the neural network, which was trained on less than 200 pictures, but still maintains a ~60% accuracy for six classes is definitely an personal achievement. Although it was both our first time using HTML, it was very fun to struggle together and through our blood, sweat, and tears, we created a basic functioning website (only slightly infested with bugs).

What we learned

For the both of us, this is the first project we had done outside of school work that has such a tight time constraint. Working as a team and coming up with a realistic project idea was some of the things we had to learn along the way. Originally we were planning to create an mobile app, which keeps track of how many items you recycled today and award you points. However, when we starting coding the app, we decided that we would not be able to finish the app within the time constraint. Instead we decided to focus on the core features of the app and integrated it into a website instead.

What's next for Recyclabot

The first feature that we really regret we can't implement is the self training neural network, due to time constraint. We didn't quite manages to create a scheduler to tells the neural network to take in the user input and trained itself. So that is definitely the first thing that needs to be done next.

Moving past that, the next step would definitely be to improve the user experience. The biggest challenge with doing that will be making the GUI more user friendly and porting it over to mobile to make it more accessible. By making it as convenient and fun as possible we hope to not only encourage people to recycle but also help informed them about what is recyclable.