Deep Recyclable Trash Sorting Using Integrated Parallel Attention.

Area of Science:

• Computer Vision
• Machine Learning
• Environmental Science
• Sustainable Development

Background:

• Manual trash sorting is inefficient and labor-intensive.
• Existing computer vision models struggle with background interference in trash image classification.
• Lack of diverse, real-world trash datasets hinders model development.

Purpose of the Study:

• To introduce the TrashIVL dataset for training and evaluating recyclable trash sorting models.
• To design a novel attention module (IPAM) to improve model robustness against background noise.
• To advance automated trash sorting using computer vision for enhanced recycling efficiency.

Main Methods:

• Development of the TrashIVL dataset with 5 (TrashIVL-5) and 12 (TrashIVL-12) classes, featuring real-life backgrounds.
• Proposal of the Integrated Parallel Attention Module (IPAM) for focusing on essential trash features.
• Integration of IPAM into a convolutional neural network (CNN) for a recyclable trash sorting system.

Main Results:

• The proposed model achieved 97.42% accuracy on the TrashIVL-5 dataset.
• The model achieved 94.08% accuracy on the more complex TrashIVL-12 dataset.
• The IPAM effectively mitigates background interference, enhancing sorting performance.

Conclusions:

• The TrashIVL dataset and IPAM provide a robust solution for automated recyclable trash sorting.
• This work demonstrates the significant potential of computer vision in environmental protection.
• The developed system contributes to improved recycling efficiency and sustainable development.