AI-based plastic waste sorting method utilizing object detection models for enhanced classification.

Area of Science:

• Environmental Science
• Computer Science
• Materials Science

Background:

• China's plastic waste export ban highlights the need for advanced domestic recycling solutions.
• Effective sorting is critical for efficient plastic waste management and resource recovery.

Purpose of the Study:

• To evaluate the performance of Mask R-CNN and YOLO v8 for plastic waste sorting.
• To compare these AI models based on key performance metrics and inference speed.

Main Methods:

• Utilized Mask R-CNN and YOLO v8 for plastic waste identification and segmentation.
• Performed hyperparameter tuning using grid search for model optimization.
• Evaluated models on accuracy, mean average precision (mAP), precision, recall, F1 score, and inference time.

Main Results:

• Mask R-CNN achieved high accuracy (0.912) and mAP (0.911) for detailed segmentation but had slower inference (200-350 ms).
• YOLO v8 demonstrated excellent mAP (0.922) and faster inference (80-160 ms), suitable for real-time applications, with an accuracy of 0.867.
• Model performance varied based on task requirements, balancing segmentation detail with processing speed.

Conclusions:

• Both Mask R-CNN and YOLO v8 show significant potential for enhancing plastic waste sorting.
• The selection of an AI model should align with specific recycling application needs, prioritizing either detailed analysis or rapid processing.
• This research provides valuable insights for optimizing automated sorting systems in the plastic recycling industry.