Machine learning in inorganic waste classification

Abstract

Waste classification and processing are important issues in promoting a more sustainable economy and friendly living environment. Smart technology applications could change how waste could be handled and recycled to increase productivity and reduce the risks of this manual work. Several studies have applied different methods of deep learning for waste classification have been proposed. Most studies have to combine many methods to improve the deep learning model to become the most optimal. In this research, I focus on three groups of typical recyclables in Vietnam and build a model which was applied to a popular convolutional neural network CNN (Resnet50) by using the smartest approaches. By using pre-trained models by transfer learning method, and applying SGD optimizer instead of Adam optimizer, the results are obtained with relatively high accuracy of 87.05%. This could show that our model will partially help convert manual processing to automated system applications.