Real-time monitoring system of crushed waste plastic particles using deep learning based object detection for particle size distribution monitoring

This study developed a real-time monitoring system for the particle size distribution (PSD) of crushed waste plastics generated during the comminution stage of recycling. A deep learning-based object detection model was trained using four major post-consumer plastics (PET, PP, PE, and PS), achieving a mean Average Precision at an IoU threshold of 0.50 (mAP50) of 0.86, along with a recall of 0.83 and a precision of 0.99. A region of interest (ROI)-based object tracking pipeline with an additional false-positive class was implemented, resulting in an F1 score above 0.95. Among the evaluated size estimation approaches, the minAreaRect-based M3 method achieved < 4% error in D10, D50, and D90 estimation at 53 frames per second, enabling accurate real-time PSD analysis. Based on this model, feedback-guided monitoring framework was proposed to support dynamic adjustment of operating parameters when D50 deviated from a predefined range. This approach improves PSD uniformity, enhances downstream separation efficiency, and reduces energy consumption, demonstrating strong potential for application to diverse material types, shapes, and large-scale resource recovery processes.