Yolo-based real-time recyclable waste detection for reverse vending machine

Abstract

The widespread use of waste in daily life has led to severe problems related to waste management. One of the systems that offers a significant solution to these problems is the Reverse Vending Machine. Deep learning has recently attracted interest as an alternative computational method to solve various waste classification challenges. Many researchers have focused on this area, yielding significant research results in recent years. Reverse Vending Machine is an interactive platform that can increase recycling activities by rewarding users. The vending machine should be designed with a material identification module that can recognize different types of recyclable materials to reward the user accordingly. The Yolov8 algorithm is used in the system developed to detect objects in real-time. The model was trained using the TrashNet dataset for reverse vending machines, similar applications, and a study-specific dataset. A waste detection framework based on object detection is proposed to identify seven types of recyclable materials. The training was performed in about 14.5 hours using a Tesla T4 graphics processing unit with 5716 images. Then, application metrics such as F1-score, P, R, and PR curves were calculated to evaluate the training and test performances of the model. To compare the model with different algorithms, the model was trained with algorithms such as YOLOv9, YOLOv8, YOLOv7 and  YOLOv5, then it was compared with other models. As a result, this study sheds a positive light on the potential of the Reverse vending machine, one of the alternatives for recycling and waste management, to develop high-accuracy systems.