Machine Learning For The Classification And Separation Of E-Waste

Abstract

The amount of global e-waste is growing at a rapid rate and is projected to increase to 74.7 Mt by 2030. However, according to a recent United Nation’s study in 2019, the collection and recycle rate of e-waste is only 17.4%. One of the most challenging barriers in large-scale e-waste recycling is the labor-intensive sorting, dismantling, and hazardous removal process, which incentivized the illegal transfer of large amounts of e-waste from developed countries to Asian and African countries. To protect the environment and the health of workers, an automated method for sorting and separation of e-waste is urgently needed. In this project, a convolutional neural network (CNN) image-recognition algorithm was developed to classify e-waste into different categories with high accuracy. An image database of four different classes of e-waste was created, and a demonstration setup was established. The pre-process of image data, the selection of hyper-parameters and the accuracy of the CNN model were discussed. The developed CNN model exhibited a training accuracy of 96.9% and a validation accuracy of 93.9%. The use of different image sizes, data augmentation by rotation, and background removal were experimented to improve the model performance. Planned future work includes expanding the database to add more diversified classes of e-waste and exploring the creation of a conveyor-belt-based demonstration.