AI-Powered Garbage Classification: A Deep Learning Approach for Smart Waste Management

Abstract The escalating volume of municipal solid waste (MSW) due to rapid urbanization and population growth poses severe challenges to sustainable waste management. Manual sorting is labor-intensive, inconsistent, and hazardous. This paper proposes an automated garbage classification system using deep learning with transfer learning on the ResNet-50 architecture. The model is trained on the RealWaste dataset (4,752 real-world images across nine categories: Cardboard, Food Organics, Glass, Metal, Miscellaneous Trash, Paper, Plastic, Textile Trash, Vegetation) and achieves a test accuracy of 96.36%. A simple Tkinter-based graphical user interface (GUI) demonstrates real-time classification capability. Results show superior performance compared to baseline CNNs (~72%) and traditional methods (60–70%), offering a practical solution for smart bins, conveyor systems, and municipal frameworks. This work supports enhanced recycling efficiency, reduced environmental impact, and alignment with Sustainable Development Goals 11 and 12. Keywords: Waste classification, deep learning, convolutional neural networks, transfer learning, smart waste management, ResNet-50, RealWaste dataset