DeepWaste: Applying Deep Learning to Waste Classification for a Sustainable Planet

Abstract:

Every year the world generates over 2 billion tons of solid waste. Even though 2/3rds of this waste is recyclable, more than 75% of it ends up in our landfills. These landfills generate over a billion metric tons of CO2 equivalent greenhouse gases, contributing nearly as much to global warming as all the cars on the US roads. These catastrophic environmental outcomes largely stem from human confusion in the identification and correct disposal of waste into waste bins. In fact, 9 out of 10 people say they would recycle and compost more if it was easier. Because of this, there is an urgent need for a quick, accurate, and low-cost method that is available for everyone.

We propose DeepWaste, an easy to use mobile app that utilizes optimized deep learning techniques to accurately classify waste items into trash, recycling, and compost. The best model, a deep learning residual neural network with 50 layers, achieves an average precision of 0.93 on the test set and is deployed into a mobile app where it can perform real-time image classification to provide instantaneous feedback to users within milliseconds. The mobile app allows users to upload new images which are stored in a big data NoSQL database and are used to continuously train and improve the model. DeepWaste demonstrates the potential machine learning holds in mitigating climate changing – if DeepWaste can even reduce landfill waste by 10%, it will be equivalent to removing over 65 million gasoline-burning passenger vehicles from the road.