<![CDATA[Microplastic pollution in aquatic ecosystems poses a serious environmental and public health risk, requiring effective and scalable monitoring technologies. Current detection methods, which rely on manual microscopy and spectroscopic verification, are labor-intensive, time-consuming, and unsuitable for large-scale assessments. While deep learning offers a potential alternative, current approaches are often limited by dependence on non-public datasets and a lack of model interpretability. This paper presents an automated, transparent, and repeatable deep learning system for microplastic identification based on advanced YOLO (You Only Look Once) architectures. The proposed system utilizes and evaluates YOLOv8 and YOLOv11 models on a consolidated public dataset of microplastic images, using extensive data augmentation to enhance model robustness. Results show that the YOLOv11 model achieves a state-of-the-art mean Average Precision (mAP@50) of 94.7%, significantly outperforming the YOLOv8 baseline at 89.5%. Additionally, implementing Explainable AI (XAI) techniques, particularly Eigen-CAM, provides vital visual validation of the model's decision-making process by highlighting microplastic features, thereby improving interpretability and confidence. This study offers a repeatable, highly accurate, and transparent detection framework suitable for automated environmental monitoring. The findings demonstrate that Transformer-based object detection models combined with XAI can significantly enhance microplastic pollution assessment, supporting more effective monitoring and mitigation of aquatic pollution.]]>
Copyrights © 2025
