Municipal Plastic Detection and Classification in Real Time using YOLOv9 and Custom CNN

Abstract

Plastic waste management is a critical global issue, with over 380 million tons of plastic produced annually, much of which pollutes the environment. Manual sorting of municipal plastic waste is labor-intensive and costly, creating the need for automated solutions. This paper presents a real-time system that detects and classifies plastic waste using deep learning. The system integrates YOLOv9 for detecting plastic items and a custom Convolutional Neural Network (CNN) for classifying them into four categories: Polyethylene Terephthalate (PET), Polypropylene (PP), High-Density Polyethylene (HDPE), and Polystyrene (PS). Our experiments show that YOLOv9, evaluated as an object detector, achieved a precision of 95.77%, a recall of 97.04%, and mAP@50–95 of 90.92%, while the CNN, evaluated as a classifier, achieved a precision of 98.41%, a recall of 98.42%, and an F1-score of 98.40%, demonstrating strong classification performance. These results indicate that the developed system provides an effective, scalable, and cost-efficient approach for automating plastic waste sorting, supporting improved recycling and sustainable waste management.