Recycle Waste Detection and Classification Model Using YOLO-V8 for Real-time Waste Management

• Published in: 6th IEEE International Conference on Artificial Intelligence in Engineering and Technology, IICAIET 2024
• Main Author: Rastari M.A.M.; Roslan R.; Hamzah R.; Ibrahim Teo N.H.; Shahbudin F.E.; Samah K.A.F.A.
• Format: Conference paper
• Language: English
• Published: Institute of Electrical and Electronics Engineers Inc. 2024
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85209667091&doi=10.1109%2fIICAIET62352.2024.10730703&partnerID=40&md5=e14baab400f89a7f185f232b35363be3

The waste management organisations face significant challenges in effectively identifying and classifying waste materials. As of 2023, the world generates approximately 2.1 billion tonnes of municipal solid waste annually, with projections estimating this will increase to 3.8 billion tonnes by 2050. However, traditional manual methods for waste segregation are time-consuming, labour-intensive, and prone to errors. Numerous artificial intelligence applications in waste management can greatly benefit from the use of technology to streamline and improve the waste detection and classification process. Therefore, this paper presents and studies the recycle waste management model that utilizes the You Only Look Once Version 8 (YOLO-v8) object detection model for real-time waste identification and classification. All the image datasets have been collected and captured in the Malacca and Selangor areas. Additional dataset acquired from the Garbage Classification Dataset. To evaluate the quantitative predictive performance of the proposed model, model summary, confusion matrix, accuracy, precision, recall, and F1-score are computed with random-splitting and manual-splitting of train-test image dataset with the ratio of 70: 30, 80: 20, and 90: 10. The model tested on 4039 images with four types of recycled waste that are paper, glass, metal, and plastic images. Then, pre-processing with data augmentation to evaluate as experimented with 10057 images. The experimental results showed the ratio of train-test for accuracy of 97.63%, precision of 95.3%, recall of 93.03%, and F1-score of 97.63%. This model showed that the predictive model trained and tested on real-time data from laboratory findings can be used to predict four types of recycled waste, and can be extended to other types of waste and applications in other domains. This proposed model has great significance for prediction study for recycling waste detection and classification. This model demonstrated its potential for future use in a wide range of other classification prediction applications. © 2024 IEEE.