<![CDATA[Improper waste management remains a significant global challenge, resulting in severe environmental and health impacts. Existing classification systems were designed and studied on large deep learning models, which are computationally expensive and not well-suited for embedded systems. To overcome this challenge, this study introduces a lightweight NasNet Mobile architecture that was trained using a three-stage learning framework. The framework employs transfer learning, fine-tuning, and hyperparameter optimisation to improve the model’s performance and generalisation capabilities progressively. To validate the proposed approach, experimental evaluations were conducted on TrashNet and Garbage Classification datasets. The model achieved an accuracy of 91.25% on the TrashNet dataset and 94.85% on the Garbage Classification dataset using the optimal hyperparameter set obtained through the random search technique. These results indicate that the proposed strategy effectively adapts to varying data distributions and outperforms popular Convolutional Neural Network (CNN) architectures, such as VGG-16, ResNet, AlexNet, etc. Therefore, the proposed model provides a reliable foundation for developing scalable and efficient waste classification systems for environmental applications. This study contributes to a practical deep learning approach that improves classification performance while maintaining low resource requirements for sustainable waste management.]]>
Copyrights © 2025
