Detection of plastic particles in water based on deep learning

Abstract

In recent years, plastic pollution has emerged as a pressing global environmental issue. The detection and classification of plastic particles play a crucial role in pollution control and material recycling. However, traditional detection methods, such as manual counting and spectral analysis, are characterized by low efficiency, high costs, and significant complexity, making them inadequate for rapid and large-scale detection requirements. This study proposes an innovative and efficient plastic particle detection and counting scheme that integrates fluorescent dyeing technology with deep learning methodologies.

Firstly, the fluorescent dyeing method was optimized to enhance the visualization of plastic particles in microscopic images, thereby providing a robust data foundation for subsequent image analysis. Subsequently, the YOLO (You Only Look Once) object detection algorithm was employed to automate the detection and counting of plastic particles in microscopic images. Through data augmentation and model optimization, the detection accuracy and generalization capability were significantly improved. Experimental results demonstrate that the proposed scheme achieves high accuracy (mAP > 90%) and efficiency in detecting both individual and clustered particles. Compared to conventional methods, this approach substantially reduces detection time and effectively handles complex backgrounds and diverse particle characteristics.

This research not only offers a viable technical solution for the rapid detection of plastic particles but also establishes a solid technical foundation for the automated analysis of plastic pollution control and microplastics research. Future applications of this method could extend to other complex particle system detection and classification scenarios, potentially integrating multi-modal data (such as spectral analysis) to further enhance particle classification accuracy.