Enhancing Smart Waste Management through Deep Learning: A Comparative Analysis of CNN and ResNet-18 Models

Abstract

Effective waste management is vital for environmental sustainability and public health. This study applies deep learning, particularly Convolutional Neural Networks (CNNs), to develop an image-based waste classification system that identifies and categorizes waste into eight distinct classes. A user-friendly interface enables image uploads and generates tailored disposal recommendations. Two models were evaluated: a custom-designed CNN and a pre-trained ResNet-18. Both models were trained and validated using three data split ratios: 70/30, 80/20, and 90/10. The custom CNN achieved a peak accuracy of 85.1% under the 90/10 split, while ResNet-18 outperformed it across all splits, attaining up to 96.1% accuracy. The superior performance of ResNet-18 is attributed to its deeper architecture and large-scale pre-training on the ImageNet dataset. These results highlight the potential of CNN-based models, especially pre-trained architectures, for automating and enhancing smart waste management systems.

Keywords: Convolutional Neural Networks (CNN), ResNet-18, waste classification, deep learning, image recognition, smart waste management.