Implementation of DenseNet121 Architecture for Waste Type Classification

Abstract

The growing waste management problem in many parts of the world requires innovative solutions to ensure efficiency in sorting and recycling. One of the main challenges is accurate waste classification, which is often hampered by the variability in visual characteristics between waste types. As a solution, this research develops an image-based litter classification model using Deep Learning DenseNet architecture. The model is designed to address the need for automated waste sorting by classifying waste into ten different categories, using diverse training datasets. The results of this study showed that the model achieved an overall accuracy rate of 93%, with an excellent ability to identify and classify specific materials such as batteries, biological materials, and brown glass. Despite some challenges in metal and plastic classification, these results confirm the great potential of using Deep Learning technology in waste management systems to improve sorting processes and increase recycling efficiency.