1 of 16

Helping save the earth by reusing and recycling one step at a time

Earth

Buddy

2 of 16

Not enough people know about efficient ways to recycle

Problem:

3 of 16

Introducing

Finding easy and fun solutions to save the earth!

Earth

Buddy

4 of 16

Part 1:

Aggregate Data

5 of 16

Dataset

By using an open source dataset, we resized and labelled all data to allow for more efficient processing by our model

~500 images of each of the various materials were used for training

6 of 16

Part 2: Develop Machine Learning Model

7 of 16

ML Model

The model is based on a 34 layer resnet34 CNN

Due to our lack of computational resources, training the model took 20 epochs, each taking up 45 minutes, which was the most time we were able to give resulting in a final successful identification rate of materials at 92.6%. This could be increased through more training time which is outside the scope of this hackathon.

Because our images were taken in a controlled environment, in normal use cases we would be able to collect and train on more and more data resulting in a more accurate material identifier

8 of 16

Part 3:

Predicting material based on photo

9 of 16

Trained Model

Using the previously trained ML model, we were able to achieve a very low validation error with low loss

ML Model

Photo

Confidence rating

prediction/actual/loss/probability

These top images were the most poorly performed on, mostly because they were overexposed which means it is not the fault of the model

- The model also has lower performance on glass materials but this would improve with use over time

10 of 16

How do you use Earth Buddy?

You first choose if you want to learn about recycling or reusing with the respective buddies

Using ML, they detect the material of the object you take a picture

You take a picture of the object you want to learn about

Using google maps as an API, buddy will find the closest recycling center for the user to go recycle!

Searches the pinterest for DIY ideas to reuse the objects you want to recycle!

11 of 16

What now?

This can be implemented to to identify objects with their respective materials and provide corresponding ways to recycle or DIY this object.

12 of 16

With this invention we hope to bring the knowledge of how to recycle and reuse closer to the general public and provide an easy way for them to join the green movement!

13 of 16

With Earth Buddy, we can help save the earth and refresh our world from waste pollution.

Thank you!

Earth

Buddy

14 of 16

Resources

15 of 16

All of our components are hand-drawn:)

16 of 16

ML Modeling

Frontend