smart trashcan
smart trashcan that can recognize and sort 6 different types of waste
At Cornell’s 24-hour Make-a-thon, my teammates (Tiantian Li and Raymond Chang) and I built a smart trashcan that recognizes and sorts 6 different types of waste to sustainability reduce landfill waste and encourage recycling, winning 3rd place.
On the software side, we researched various datasets for recyclable image classification and decided to use a combination of TrashNet (from Github) and Waste Classification Data (from Kaggle). We used a pretrained VGG-16 deep convolutional neural network, with our own custom classification head. Our model achieved an accuracy of 85% overall. We did some fine tuning and biased predictions towards ‘trash’, achieving 95% accuracy on the trash category in the end.
On the hardware side, with limited access to hardware and fabrication, we were able to make a concept trashcan that displays the trashcan’s overall goal. The trashcan is fabricated with 3d printed, laser cut parts, and fasteners. The trashcan 6 different waste bins with one neutral position. When the trashcan recognizes what type of trash it is, the servo turns to open up the corresponding trash bin (trash, cardboard, glass, metal, paper, and plastic).
Further software and hardware implementations can be found in the video and the presentation.