<![CDATA[Operational efficiency of municipal solid waste management is critically hampered by manual sorting, a process that is slow, labor-intensive, and poses occupational health risks. This study investigates the application of deep learning to automate this task, specifically by developing and accessing a classification model using Residual Network (ResNet-50) architecture. Utilizing a public dataset of 24,705 images of organic and recyclable waste, we implemented a transfer learning protocol. The pre-trained ResNet-50 model, with weights initialized from ImageNet, was fine-tuned for this binary classification problem. To ensure model robustness and prevent overfitting on the specialized dataset, a comprehensive data augmentation strategy was applied, expanding the training set to 207,522 images. Upon evaluation with a dedicated test set of 6,177 that was previously unseen, the model demonstrated high evaluation metrics, achieving an all-around classification accuracy of 95%. Another key evaluation metrics, like precision, recall (sensitivity), specificity, and F1-score, corroborated the result with a weighted average of 0.95. These findings establish that a transfer-learned ResNet-50 model provides a highly effective and reliable solution for automated solid waste classification, offering a technologically robust foundation for next-generation recycling and waste management systems.]]>
Copyrights © 2026
