Enhancing trash classification in smart cities using federated deep learning.

Area of Science:

• Computer Science
• Environmental Science
• Artificial Intelligence

Background:

• Smart cities require efficient waste management for environmental sustainability.
• Effective trash classification is key to optimizing solid waste management systems.
• Deep learning offers potential for advanced waste classification solutions.

Purpose of the Study:

• To compare deep learning models for trash classification in smart cities.
• To identify the most effective convolutional neural network (CNN) model for this task.
• To explore a federated learning framework for enhancing trash detection.

Main Methods:

• Comparative analysis of ten deep neural network models using PyTorch.
• Experiments conducted on the TrashBox dataset.
• Evaluation of model performance based on training, validation, and test accuracies.

Main Results:

• The ResNext-101 model demonstrated superior performance across all accuracy metrics.
• Consistent high accuracy achieved by ResNext-101 compared to other tested models.
• The study identified ResNext-101 as a highly effective CNN for trash classification.

Conclusions:

• CNN-based trash classification significantly advances smart city waste management.
• ResNext-101 is a promising model for efficient and accurate trash detection.
• Federated learning offers a pathway to further optimize combined CNN model performance.