Inspiration

When we heard about the sustainability challenge, we were very keen to make something that could have a real impact for our planet. Our idea was inspired on "reverse vending machines" where you give back water bottles to a machine that will recycle them. Our project consists of expanding on this idea with the ultimate goal of creating an automated process which can sort different types of rubbish rapidly, reliably and, crucially, with no human input required.

What it does

Our project simulates this hardware implementation by having an algorithm that, given an image of an item, detects what category of waste that it belong to: plastic, cardboard, metal, etc. We also built a website to promote our project. Ideally, the website will contain a section where one can upload an image which will be processed by the algorithm. Then, the waste category will be returned.

How I built it

Using a preexisting data-set from kaggle and library fastai (built on pytorch), we developed and trained a convolutional neural network to determine the waste category of an image. In addition, we built a website using a modified html template to promote our project.

Challenges I ran into

The main challenge we ran into was being unable to install pytorch on our lab machines due to storage issues. In the end, we resolved this issue by downloading it on our mentor's account. Additionally, we did not manage to link our website with the machine learning implementation in time.

Accomplishments that I'm proud of

Our biggest accomplishments were being able to use a data-set to train a model with 94% accuracy and to modify a html template to build our own website.

What I learned

We learnt a wide variety of new skills such as html, python programming, and the main aspects of machine learning.

What's next for What's that rubbish?

The main goal of this project would be to have a hardware implementation of this machine. Then, when a user inserts a piece of rubbish into the machine, it will automatically classify it and recycle it. Firstly, we would begin by linking the website to the algorithm and then build a hardware implementation.

Built With

Share this project:

Updates