E-container

Inspiration

Do you know what happens when you put the wrong thing in any garbage container or recycling bin? Sorting is the most crucial recycling cycle process and the most expensive and complicated.

At least the 30% of objects sorted in any category should have never been there. This causes severe problems with the contamination of the different things that have to be sorted manually or by a machine. For example, contaminated plastic bags can wrap around machinery and could cost over $10,000 to repair.

The environment is also the victim of this. If a recyclable object can't be sorted, it won't be used in any future product that should use recyclable components, forcing to waste more natural resources.

We can solve all of that through technology, and now more than ever, we can be proud of our team philosophy: It's time for change!

What it does

Through any type of camera on any platform or device, E-container can recognize different types of objects and recommend to the users what garbage container they should use.

During the ToHacks Hackathon, we have been able to train our ML model with seven main categories:

Plastic (plastic bags, plastic bottles, and plastic bricks)
Paper
Cardboard
Glass
Local waste facility (microwave, computer)
Batteries
Waste

How we built it

How it works?

E-container must be an application available on all platforms and devices, so we decided to develop our app using Flutter (with Dart). With the help of TensorFlow Lite, we were able to train a Machine Learning (ML) model of image classification using the power of the Google Collab (with GPU) instances.

We have different steps involved:

1- Build a new dataset: we started building an entirely new dataset for garbage classification, optimized for E-container.

2- Train a TensorFlow Lite model: we train the model using our dataset, and the images from the feedback users submit through the app and stored in CockroachDB.

3- Recognition: our app can recognize objects such as plastic bags, glass, electronic components, etc. The most reliable result from our ML model will determine which garbage container the user should use.

4- User feedback: if any object recognition it's not ok, any user can submit with a simple button press an image of the object that E-container does not recognize correctly. That information will be stored in CockroachDB for further ML training.

What's next for E-container

Our most important and exciting challenge is improving our ML model and knowledge and adapting E-container to different countries and regulations. It will also be a good idea to include some gamification or reward system, which would be very interesting for city councils or governments looking for ways to encourage citizens to be a crucial part of the recycling cycle.