Deep Learning Methods for Waste Classification

Abstract

Waste could be reduced, reused and recycled. Classifying waste and recycling play important roles in converting waste into valuable materials that help conserve land reduce pollution and optimize resource utilization. Effective waste management is vital for resource conservation, environmental protection, and sustainable human progress. However, there are difficulties in classifying and identifying recyclable materials due to the complex and diverse nature of waste combined with limited data on waste management. These challenges pose obstacles to the effectiveness of research, in this field.

The development of deep learning has promoted the advancement of pattern classification and visual object detection. Applying computer vision to waste classification and exploring high-efficiency, low-cost, and automated waste classification methods have important sustainability implications for society and the ecological environment. Taken the current limitations of waste classification into account, this project makes use of deep learning methods to improve waste classification from several different perspectives. Firstly, we build two comprehensive and rich waste datasets, including four waste categories, namely recyclable waste, wet waste, hazardous waste, and dry waste. Afterwards, we have made innovations in data augmentation and proposed NUNI, a non-uniform data augmentation method, to improve the accuracy of waste classification. Finally, we propose a semi-supervised learning deep learning framework, called CISO, and utilize it on the waste classification task. We evaluate our models and provide the relevant comparisons.