Deep Learning with YOLOv8 for the Classification of Solid Waste According to the Peruvian Technical Standard 900.058:2019

Del Piero C.C.A.; Manuel C.N.R.; Santy P.T.A. · 2024 18th National Meeting on Optics and the 9th Andean and Caribbean Conference on Optics and its Applications, ENO-CANCOA 2024 - Conference Proceedings (2024) · DOI: 10.1109/eno-cancoa61307.2024.10751320

Resumen

Poor management of solid waste presents a pressing global challenge, leading to substantial social and environmental consequences. This study introduces a waste detection and classification system based on the deep learning model YOLOv8 and the Peruvian Technical Standard 900.058:2019. Techniques such as data augmentation by Roboflow and GroundingDINO-assisted labeling were utilized to construct a robust and representative dataset comprising approximately 4000 images of debris. Following rigorous training of 100 epochs, the model achieved an average accuracy of 95.8% and a recall of 0.933 in detecting waste categories including plastic, metal, glass, paper, cardboard, and bio-waste. Additionally, a graphical interface was developed to enable real-time execution of the model, yielding an average accuracy of 89.42%. These findings underscore the efficacy of the proposed approach for waste classification, aligning with technical standards and contributing to enhanced waste management practices. Nonetheless, certain areas for further research and improvement are also identified.

Instituciones: Universidad Nacional del Centro del Perú