2026-05-13T16:21:10.333Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11369 2025-04-08T00:17:04Z 3

Exploring the potential of lightweight computer vision YOLOv8 models for effective waste classification and management 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11369 https://www.sciencedirect.com/science/article/abs/pii/B9780443273742000108 A. Tasic L. Jovanovic N. Bacanin M. Zivkovic S. Kozakijevic M. Antonijevic M. Popovic This study delves into the practical application of YOLOv8 models in waste item classification through computer vision, a critical pursuit in the realm of waste management amid escalating global waste volumes. As waste reduction becomes increasingly urgent, recycling emerges as a pivotal strategy. However, the effectiveness of recycling efforts is often impeded by challenges like waste separation, a labor-intensive and hazardous process when conducted manually. Here, the study investigates the potential of leveraging computer vision to streamline trash separation processes. To address computational constraints, the study adopts lightweight YOLOv8 architectures, encompassing nano, medium, and small models. While each trained model demonstrates promising performance, the small model stands out for its commendable balance between computational efficiency and classification accuracy. Of particular significance is the models’ adeptness in discerning metal and glass items, given their high recyclability rates. This research lays the groundwork for more efficient waste management practices, highlighting the transformative potential of computer vision technologies in addressing pressing societal challenges. bookPart Elsevier 263 281 10.1016/B978-0-443-27374-2.00010-8 Chapter in Harnessing Automation and Machine Learning for Resource Recovery and Value Creation