WasteNet for Waste Classification

Poster

Introduction

With accelerating development in manufacturing and E-commerce, our society is now faced with an ever-growing amount of waste, much of which goes abandoned without being recycled and reused. This leads to tremendous loss of resources and even more hostile conditions for our environment, causing pollution in water, air, and soil. To solve the problem of recycling, classification becomes essential. Waste classification is a very promising but not yet widely industrialized application of deep learning. However, the current process of waste classification is largely led by human force, and its practical enforcement has proven inefficient. Therefore, deep learning comes into the picture as a potential help: using computer vision to automate the classification process will be a very promising application of neural networks for making this irksome but essential issue approachable. For our project, we draw inspiration from various network architectures including DenseNet, SparseNet, and Multipath-DenseNet, and proposed a new model, WasteNet, which has improved performance compared to the state-of-the-art convolutional neural networks.