Garbage detection and classification using a new deep learning-based machine vision system as a tool for sustainable...

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Effective waste recycling is crucial for environmental pollution management.
• Garbage classification accuracy directly impacts recycling efficiency.
• High labor costs and limited capacity hinder current waste sorting methods.

Purpose of the Study:

• To develop a machine vision system for automated garbage detection and classification.
• To reduce labor costs and increase garbage classification capacity using deep learning.
• To enable real-time garbage sorting on edge devices for sustainable waste recycling.

Main Methods:

• An improved MobileNetV2 deep learning model incorporating an attention mechanism was developed.
• Transfer learning with pre-trained weights enhanced model generalization.
• Principal Component Analysis (PCA) reduced model dimensionality for edge deployment.

Main Results:

• The model achieved 90.7% classification accuracy on the Huawei Cloud dataset.
• Average inference time was 600 ms on a Raspberry Pi 4B.
• A prototype demonstrated 89.26% accuracy in real-world garbage identification.

Conclusions:

• The developed deep learning model significantly improves garbage classification accuracy and efficiency.
• The machine vision system offers a viable solution for automated waste sorting.
• The garbage sorting prototype serves as an effective tool for sustainable waste recycling.