A Material-Based Litter Dataset Preparation For User-Friendly Marine Environment Monitoring

Marine litter detection remains a critical environmental challenge. Building on our previous evaluation of the PlastOPol dataset and object detection models, this research addresses key gaps in real-time detection and dataset coverage. While our earlier study demonstrated YOLOv5’s superior performance on low-power devices, it revealed limitations in existing datasets’ material-based classifications. We address these gaps by introducing a comprehensive marine litter dataset with material-type annotations and evaluating YOLOv11 for real-time detection on lowpower devices. Our advancements significantly enhance environmental monitoring capabilities and support citizen science initiatives.