STEW: Steps To End Wastes

STEW: Steps To End Wastes is a published application that suggest users to bring eco-friendly item instead of single use plastics when they are going out.

Idea

Everytime the users going out, the application will suggest eco-friendly item to bring along. After prepared all the items, the users will submit the picture of the items and the Neural Networks will detect the objects and awards them green points

Applications

We have three main activities:

Start Activity -> Suggestion Activity -> Classification Activity

Start Activity: Register account on Firebase Database

Suggestion Activity: Choose the destination and provide suggestions

Classification Activity: User uploads images of item then getting rewards

Start Activity

Classify Activity

Transfer learning: Xception model

I implemented transfer learning using Google's Xception model, in which I only change the model's output nodes

Database

We linked our Android Studio app with Google Firebase Database so that the program could add the points after the user validates the items they bring along with

Deploy model onto mobile

From the current model on Tensorflow, I converted the model to Tensorflow lite version so that I can deploy it on mobile platforms.

On mobile, *.tflite model required input to be ByteBuffer type, so in order to retrieve the prediction, I've converted the image captured from Bitmap to ByteBuffer

Community and results

Model results

This graph depicts how the model was trained throughout 800 epocs.

Community

We created a community named STEW, where people could share their environment's saving results.

During the demo stage, we have collected the total number of eco-friendly used during 9 days.

We also held a ranking system while in the demo stage and giving out prizes for people that saving the environment the most

Back