Microplastic Detection in Aquatic Environments Using YOLOv11 and Explainable AI Techniques

Microplastics present a major threat to aquatic ecosystems, making the development of accurate detection methods essential for effective monitoring. This study proposes a novel approach for detecting microplastics in water using YOLOv11, an advanced deep learning model. YOLOv11 achieved impressive performance, with a detection accuracy of 87.5%, surpassing its predecessors: YOLOv8, YOLOv9, and YOLOv10. The model utilizes a fully transformer-based backbone, dynamic multi-scale feature aggregation, and an anchor-free design, enabling it to efficiently detect small and irregularly shaped objects in complex aquatic environments. To enhance interpretability, Explainable AI techniques specifically CAM and Eigen-CAM were used to visualize the regions influencing the model's predictions, with YOLOv11 accurately highlighting the relevant microplastic areas and confirming its superiority as the most effective model for microplastic detection. The dataset, which was created from microscopy images, was augmented through methods such as rotation, flipping, and brightness adjustments, enhancing its generalizability in diverse conditions. This research presents a scalable solution for automated water quality monitoring, effectively addressing challenges related to low-resolution images, noisy backgrounds, and the varied characteristics of microplastics. However, limitations include the diversity of the dataset and the computational demands for real-time deployment. Future work will focus on integrating multi-modal data to enhance both detection capabilities and scalability. This research lays the foundation for automated and scalable microplastic detection systems, enabling more effective control and long-term monitoring of microplastic pollution in aquatic environments.