Edge-Optimized Deep Learning Models for Automated Waste Classification

This paper presents a detailed comparative analysis of four deep learning models—YOLOv8, YOLOv8-Nano, VGG16, and ResNet-18—for edge-based garbage classification, aiming to identify optimal solutions for real-time, resource-efficient waste monitoring. A large dataset of labeled images is utilized to evaluate these models across key metrics such as accuracy, inference speed, model size, and computational efficiency. YOLOv8 demonstrates exceptional real-time detection capabilities, while the YOLOv8-Nano version, optimized for edge devices, offers significant improvements in speed and resource usage. VGG16 and ResNet-18 provide strong classification accuracy but demand higher computational resources. Quantitatively, YOLOv8 achieves the highest accuracy of 95.1% and an F1-score of 95.8%, while YOLOv8-Nano balances speed and performance with an F1-score of 93.1% and accuracy of 92.5%. This research highlights the trade-offs between detection speed, accuracy, and computational efficiency, offering practical insights for deploying deep learning models on edge devices in sustainable waste management systems.