Inspiration
Whenever we go to the dining hall or food court, there is always excess waste in the trash and perfectly good food always goes to waste. We thought that we could address this problem by creating a CNN model that could recognize and organize which trash items are organic or recyclable. This would help with reducing waste, which helps the environment. We also thought that using the Brooklyn Waste Dataset, we can create a dashboard that could be used to explain to people the horrible effects of excess trash.
What it does
Our Waste Classifier inputs images of various materials and outputs whether it is organic or recyclable. The Dashboard signifies the contents of trash and we hope that our ML model can be used to create data just like the Brooklyn Waste dataset, which is what we built the dashboard using Microsoft Power BI & Machine Learning Studio.
How we built it
We utilized a convolutional neural network model from Tensorflow, which takes in a dataset of pictures that are broken down into NumPy Arrays. This involves creating several layers (Convolutional, Max Pooling, Flatten, Dense), in order to manipulate each photo into a format that can then be used to identify whether or not it is organic or recyclable.
Challenges we ran into
We had to learn CNN from scratch, and it took a very long time to train to model with the image dataset, so we only got to run it a limited number of times. We attempted to train the model on Microsoft Azure Machine Learning Workspace, however it took as long to run as a Jupyter Notebook, but it had less memory.
Accomplishments that we're proud of
We're proud that our Waste Classifier model has great accuracy, and that we learned how to use Power BI. Got a glimpse at the potential machine learning has for making a difference when it come to solving issues like sustainability, we were able gather information from trash at a small scale with limited data.
Future Implementation: With proper support we will be able to scale our project to other cities and gather more information.
Log in or sign up for Devpost to join the conversation.