At a time where pollution is at its highest, it is more important than ever for us to properly dispose of our waste.
What it does
EasyDispose classifies different wastes into recyclables, organic waste, or non-recyclables.
How we built it
EasyDispose was built using TensorFlow with a convolutional neural network that trained with a wide range of waste images. OpenCV was used for video processing allowing users to show waste to be classified.
Challenges we ran into
One of the main challenges was overfitting the data. Initially, the accuracy of the neural network was only around 60%. However, by reducing filters in the network and adding a dropout layer, which essentially removes some inputs in the hidden layers to help against overfitting, I was able to increase the accuracy up to 70%.
Accomplishments that we're proud of
Despite not having access to a GPU, I was able to train the model and achieve an accuracy of 70%, which isn't ideal but it is good enough to appropriately classify a wide range of wastes.
What we learned
I learned how use OpenCV for video processing and implement TensorFlow with it in order to allow the model to directly work of off the video frames.
What's next for EasyDispose
I want to continue to improve EasyDispose by allowing users to be able to determine the material of many wastes, such as if it is plastic, steel, etc. I also want to train the neural network with more data to improve its accuracy even further for a wider range of waste material.