Classification of recyclable waste using deep learning architectures

Abstract

Managing waste in big cities is a big problem. Wastes are dangerous in terms of causing environmental pollution and affecting human health. In particular, solid wastes such as glass and plastic do not dissolve in the soil for a long time and pollute the environment. By recycling such solid wastes, the surrounding waste can be reduced. Therefore, it is important to classify waste and to recycle the separated waste. In this study, a data set consisting of 22500 waste images was used. The data set contains color image data with a size of 227 x 227 pixels. The data used in the study are divided into two as organic and recyclable waste. This study proposes a deep learning-based system for classifying waste. With such a system, wastes can be classified and recycled. The data was trained with the ResNet 50 architecture and the CNN architecture created to classify waste, and accuracy rates were compared. The CNN architecture created to classify waste is more successful for this data set with an accuracy rate of 91.84%.