Comparative Analysis of ResNet-50 and VGG16 Architecture Accuracy in Garbage Classification System

Abstract

Population growth and urbanization have resulted in an exponential increase in waste generation, causing serious environmental and health risks. Garbage classification is essential to optimize the recycling process and minimize waste in landfills. In particular, Convolutional Neural Networks (CNN) and Deep Learning, has shown effectiveness for image classification systems such as waste sorting. This research addresses the gap in comparative analysis of CNN architectures for garbage classification by comparing the performance of VGG16 and ResNet-50. This study's objective is to identify the most effective architecture for categorizing six different categories of garbage: cardboard, glass, metal, paper, plastic, and trash. Using a dataset of 2,467 photos, the models were trained, validated, and tested using improved preprocessing and data augmentation techniques. The results showed that VGG16 obtained slightly greater accuracy (97%) than ResNet-50 (96%), indicating that VGG16 could be a better architecture for garbage classification systems. This study helps further development of automated waste sorting systems for recycling management, paving the way for more sustainable waste solutions. Hope for future research, this study can help in expanding the dataset, then using other architectures to improve the accuracy of the model, and help people to process garbage according to the type.