Performance Evaluation of YOLOv8 and YOLOv11 for River Waste Detection

River pollution caused by plastic waste requires an effective and automated monitoring solution. This study proposes an automated waste detection system in river environments by implementing and comparing two deep learning-based object detection models, YOLOv8 and YOLO11. The dataset used is the River Trash dataset (Version 3) from Roboflow, consisting of 415 original images augmented to 1,281 training and 367 validation images across five waste categories: non-plastic, plastic bag, plastic bottle, plastic others, and plastic wrapper sachet. Both models were trained under identical conditions — 10 epochs, image size 640×640, batch size 16, and AdamW optimizer — to ensure a fair comparison. Performance was evaluated using Precision, Recall, mAP@50, and mAP@50-95. Results show that YOLOv8 achieved higher detection accuracy with Precision 0.880, Recall 0.889, mAP@50 0.946, and mAP@50-95 0.942, outperforming YOLO11 which recorded 0.879, 0.800, 0.910, and 0.909 respectively. However, YOLO11 demonstrated greater efficiency with fewer parameters (2.59M vs 3.01M) and faster inference speed (249ms vs 265ms), making it more suitable for edge device deployment. These findings confirm that both models are capable of detecting plastic waste in complex river environments, with YOLOv8 recommended for accuracy-critical applications and YOLO11 for resource-constrained real-time monitoring systems.