Take an image of an object using the computer's webcam. We feed it into a trained image classification model that tells us what the object is. Then we search for the object in our recyclable database and tell you the result. If it's in the database, it's recyclable!
Our trained image classification model is a convolution neural network that we trained ourselves using google photos. The biggest challenge we ran into was reliably getting the computer to recognize different objects. We first tried identifying objects based on the shape of it's edges or the aspect ratio of it's width to height, but both these methods proved unreliable and weren't scalable to more items. So neural networks were the only way to go. We should add that there were no trained models for our purposes so we had to train our own.
We are proud that our model trained successfully and that we were able to take a new input image and return a classifier. In the future, we need to improve the accuracy of our model in addition to add more classifiers, but we are pleased with our results of this hackathon.