A Study on Construction of Waste Paper Sorting Support System by Image Recognition Using Deep Learning

Abstract

Recycling paper is an important issue in waste reduction and decarbonization. Fukuoka City has been working on the sorting of used paper as part of its waste reduction policy, but small and medium-sized enterprises (SMEs) are less likely to sort used paper than large enterprises (LSEs). In this study, with the aim of improving the sorting rate by reducing the burden of sorting, we developed a prototype system that automatically supports sorting of paper using image recognition technology based on deep learning called CNN, and examined its construction. The results showed that the use of a large-scale pre-trained model with deep layers resulted in an 88% correct inference rate by image recognition. In addition, it was found that the correct inference rate for the major categories as burnable trash and used paper was over 90% for each.