Implementation of YOLO11 for accurate waste detection

Abstract

Waste management, particularly the waste sorting process, remains a major challenge in urban areas. Manual sorting is often inefficient and prone to human error, highlighting the need for an accurate and automated waste classification system. This study investigates the implementation of the YOLO11 model for detecting and classifying six waste categories—plastic, paper, glass, metal, electronics, and organic—using a bounding box-based object detection approach. The lightweight YOLO11n variant developed by Ultralytics was employed in this study. The dataset was obtained from previous research and subsequently annotated, augmented, and divided into training, validation, and testing subsets. The model was trained for 100 epochs using the default configuration on an NVIDIA GeForce RTX 4060 Ti GPU (16 GB VRAM). During the final evaluation on the unseen test dataset, the model demonstrated reliable generalization and classification performance, achieving an overall macro-average precision of 0.75, an overall recall of 0.87, and a global F1-score of 0.79. The Paper and Plastic classes showed the highest detection performance, with Plastic achieving a flawless recall of 1.00, while the Organic class remained the most challenging due to higher visual variability. These results demonstrate that YOLO11n provides an effective approach for automated waste detection and has the potential to support more efficient and accurate waste management systems.