<![CDATA[Waste management remains a critical environmental issue globally, including in Indonesia, where increasing household waste generation creates significant environmental and social challenges, particularly at the neighborhood level. In community-based Waste Banks, manual sorting processes are often inconsistent due to limited human resources and varying levels of public understanding of waste categories. This study aims to develop and evaluate a lightweight, web-based real-time waste detection and classification system to support community-level waste management. The proposed system utilizes the YOLOv8 object detection architecture implemented through the Ultralytics framework with PyTorch as the deep learning backend, integrated with OpenCV for real-time video processing and Streamlit for web-based deployment. The dataset consists of approximately 9,200 annotated images across 24 waste categories, divided into training, validation, and testing sets, with data augmentation applied to improve robustness. Model performance was evaluated using precision, recall, and mean Average Precision at IoU 0.5 (mAP@0.5). The results demonstrate high detection performance, achieving 99.5% mAP@0.5, 99.4% precision, and 100.0% recall, while maintaining stable real-time detection under varying lighting conditions. However, these results are obtained under relatively controlled dataset conditions; therefore, further evaluation in more diverse real-world environments is necessary to ensure generalization capability. The system enables multi-object detection without requiring specialized hardware, making it accessible for neighborhood-level Waste Banks and providing a practical solution for community-based waste management.]]>
