Application of Object Recognition for Plastic Waste Detection and Classification Using YOLOv3

Abstract

Object recognition is a computer vision technique to detect the semantic of objects either in digital images or videos then to identify those objects into a particular class. This intelligent technique can be used for various applications. In this study, object recognition is implemented for real-time plastic waste detection and classification using YOLOv3. Six macro plastic waste classes are proposed, namely plastic bag, plastic bottle, crushed bottle, cup, cartoon, and straw. These six categories are usually among the top of our daily plastic waste. This classification of plastics waste aims to make the sorting task more efficient both at home and recycling center. Using around 1858 images and 2000 iterations during dataset training, results show that the detection achieves a good confidence value for plastic bottle and cartoon class which is 85% and 75% consecutively. Meanwhile, straw achieves 65% and the others are between 30 and 40%. This means the algorithm can detect and classify the plastic waste correctly. However, the further review of images used in the dataset in terms of item variety, angle, lighting, and image resolution, as well as increase the iteration number during the training phase, are required to gain a higher confidence value.