Plastiview: Deep Learning Based Microplastic Detection In Water

Abstract

Microplastics are emerging as a serious threat to both aquatic ecosystems and human health. While conventional detection methods such as FTIR and Raman spectroscopy offer high accuracy, they are often expensive, time-consuming, and unsuitable for use in field settings. To address these limitations, this paper introduces Plastiview, a low-cost, real-time microplastic detection system that combines a Raspberry Pi 4B, a USB digital microscope, and the YOLOv5s deep learning model. Using OpenCV, the system processes microscope-captured images and displays the detected microplastic count on a compact LCD screen. Trained and tested on publicly available datasets, the model achieved a precision of 0.825, a recall of 0.696, and a mean Average Precision (mAP@50) of 0.748. Designed for portability and standalone use, Plastiview offers a practical solution for environmental monitoring without the need for complex lab setups. With its modular architecture, the system also opens doors for future enhancements such as cloud-based data storage, remote access, and integration of additional sensors, making it a scalable tool for real-world applications in microplastic detection.