Multilayer Hybrid Deep-Learning Method for Waste Classification and Recycling.

Area of Science:

• Artificial Intelligence
• Computer Vision
• Environmental Science

Background:

• Effective waste sorting in urban public areas is crucial for recycling and environmental sustainability.
• Current automated waste sorting methods often rely solely on visual data, limiting their accuracy.

Purpose of the Study:

• To develop and evaluate a multilayer hybrid system (MHS) for automated waste classification in urban public spaces.
• To improve the accuracy of waste sorting by integrating image and sensor data.

Main Methods:

• A multilayer hybrid system (MHS) combining a Convolutional Neural Network (CNN) for image feature extraction and a Multilayer Perceptron (MLP) for data consolidation.
• Utilizing high-resolution cameras and sensors to capture comprehensive waste data.
• Training and validation against manually labeled waste items.

Main Results:

• The MHS achieved an overall classification accuracy exceeding 90% in two distinct testing scenarios.
• The hybrid system significantly outperformed a reference CNN-based method that used only image inputs.
• Demonstrated the effectiveness of integrating multimodal data (image and sensor) for waste classification.

Conclusions:

• The proposed MHS offers a robust and accurate solution for automated waste sorting in urban environments.
• Integrating diverse data sources enhances the performance of deep learning models for waste management applications.
• This technology has the potential to significantly improve recycling rates and reduce landfill waste.