Development of An Intelligent Classification of Plastic Waste System

Abstract

Plastic waste is a severe environmental issue in Thailand, accounting for 12% of total waste, yet only 25% is recycled. Manual classification is therefore limited and prone to errors. This study applied YOLOv11, a real-time object detection model with higher accuracy and speed than its predecessors, to the classification of four major types of plastic waste: HDPE, PET, PP, and PS. A web application was developed using Flask to enable users to detect and classify plastic waste conveniently, rapidly, and accurately. The experimental results indicated that the YOLOv11 model can effectively classify four types of plastic waste (HDPE, PET, PP, and PS), achieving an average latency of 22 ms per image and a processing speed of 55 FPS. Evaluating the classification performance by plastic type, PET achieved the highest accuracy of 0.994 and precision of 0.999, followed by HDPE with an accuracy of 0.983 and precision of 0.958. PP and PS recorded accuracies of 0.958 and 0.976, and precisions of 0.943 and 0.971, respectively. In terms of recall, HDPE reached 1.0, PET 0.97, PP 0.898, and PS 0.918. The F1-scores of PET and HDPE were the highest at 0.984 and 0.978, while PS and PP achieved 0.944 and 0.92, respectively. These results demonstrated that the model can reliably and accurately classify PET and HDPE waste. The system was suitable for practical deployment in waste sorting facilities or recycling centers, integrated with IoT devices such as robotic arms or conveyor belts, enabling real-time waste separation, reducing reliance on manual labor, and enhancing overall waste management efficiency.