International Journal of Science, Engineering and TechnologyAuthors: Ashwini Gulhane, Mohammad Ghouse Mohiuddin, Mohammad Abdul Wahab, Mohammad Faheem Pasha
Abstract: The growing volume of unmanaged solid waste poses significant environmental and operational challenges in urban settings. Traditional scrap segregation is labour-intensive, error-prone, and lacks consistency, especially in regions where manual sorting dominates the recycling process. This research presents a realtime recyclable and non-recyclable material detection system powered by YOLOv5 and Convolutional Neural Networks (CNN), integrated with OpenCV for live camera-based waste identification. A custom-labelled dataset was developed, comprising diverse waste categories such as plastic, iron, cardboard, books, motors, and nonrecyclable items to address real-world scrap scenarios. The YOLOv5 model was trained and optimized on this dataset, achieving high accuracy and rapid inference speeds suitable for deployment in industrial or public recycling environments. The system processes live video feeds and detects multiple objects simultaneously with bounding boxes and class labels, demonstrating robust performance under varying lighting and background conditions. The proposed solution significantly reduces manual effort, supports realtime sorting, and provides a scalable approach for smart waste management. This research aims to bridge the gap between intelligent computer vision applications and sustainable waste handling by enabling reliable, fast, and cost-effective classification of scrap materials.
DOI: https://doi.org/10.5281/zenodo.17699446
