Automated Microplastic Detection in Environmental Samples Using YOLOv8 Nano: A Lightweight Approach for Edge Deploymen

Microplastic pollution has emerged as one of the most pressing global environmental challenges, necessitating rapid, accurate, and scalable detection methodologies. Traditional detection techniques, while precise, are labor-intensive, expensive, and ill-suited for field applications. This paper presents the implementation and evaluation of the lightweight YOLOv8 Nano (YOLOv8n) object detection model for automated microplastic identification in environmental imagery. Leveraging a publicly available Roboflow dataset comprising 3,226 training images and 928 validation images with 6,475 annotated microplastic instances, the model was trained for 10 epochs using CPU hardware in Google Colab to simulate resource-constrained environments. The optimal checkpoint achieved a precision of 0.843, recall of 0.744, mAP@50 of 0.827, and mAP@50-95 of 0.472, demonstrating competitive performance relative to computationally intensive approaches. The trained weights were successfully exported to TensorFlow Lite (TFLite) format (approximately 11.6 MB), enabling efficient edge deployment on mobile devices and embedded systems. This work contributes a practical, accessible AI-driven tool for environmental monitoring in low-resource settings, addressing critical gaps identified in recent literature reviews. The complete pipeline, from dataset preparation to edge model export, is documented to facilitate reproducibility and adaptation by researchers and practitioners.