Efficient Waste Classification using Deep Learning: A Case Study with Garbage, Paper, and Plastic Bags using the EfficientNetB3 Model

This research study presents an efficient deep learning-based approach for the classification of waste materials, specifically focusing on garbage bags, paper bags, and plastic bags. Utilizing the pretrained EfficientNetB3 model with transfer learning, the proposed system achieves high accuracy while maintaining computational efficiency. A publicly available synthetic dataset, consisting of 15,000 images equally distributed across the three classes, was used for model training and evaluation. Various preprocessing techniques, including normalization and data augmentation, were employed to enhance model generalization. The model achieved an overall accuracy of 97%, as demonstrated by strong precision, recall, and F1-scores across all categories. The confusion matrix highlighted minimal misclassifications, primarily between visually similar categories like plastic and garbage bags. This study emphasizes the utility of the EfficientNetB3 model for real-time waste classification, contributing to improved automation in waste management systems. Future research can focus on incorporating larger real-world datasets and exploring advanced model architectures or unsupervised learning techniques to further optimize performance and generalization in practical applications. This system has the potential to significantly enhance recycling processes by automating waste sorting.