Development of an artificial intelligence-based automatic plastic waste sorting system

This study describes the development and realization of an intelligent plastic waste sorting system that combines computer vision, robotic control, and a graphical user interface (GUI) for real-time monitoring. The system integrates the You Only Look Once version 8 (YOLOv8) deep learning architecture to detect and classify seven distinct plastic types: polyethylene terephthalate (PET), high-density polyethylene (HDPE), single-use plastics, single-layer plastics, multi-layer plastics, squeeze tubes, and ultra-high-temperature (UHT) boxes. Sorting actions are executed by an EEZYbotARM Mk2 robotic arm, driven by an STM32F446RE microcontroller, which performs autonomous pick-and-place operations according to the recognition output. A custom interface built with Python Qt version 5 (PyQt5) enables live video visualization, delineation of detection areas, counting of identified categories, and additional functionalities such as counter reset, selective filtering, and exporting detection records to Excel. The complete setup was tested on a conveyor-based system, confirming high detection precision, dependable robotic operation, and a user-friendly supervision interface. The trained YOLOv8 model achieves a precision of 94%, recall of 95%, and mean average precision (mAP) at Intersection over Union (IoU) thresholds 0.5–0.95 of 94% across the seven plastic waste categories, confirming the effectiveness of the proposed detection pipeline. This contribution aims to mitigate the limited advancement of plastic recycling technologies in developing regions, offering a cost-effective and practical framework for enhancing plastic waste management.