Automated Multi-Parameter Characterization of Microplastics Using Polarization Digital Holographic Microscopy

The increasing threat of microplastics (MPs) to aquatic ecosystems has become a global environmental concern. There is an urgent need for rapid characterization techniques for MPs that enable the simultaneous measurement of multiple parameters. Here, we present what is believed to be a novel method that combines digital holography and polarization imaging to extract multi-dimensional features, including morphology and polarization, from a single holographic image. A deep learning pipeline is then proposed for automated, high-throughput particle segmentation and classification from the optical information. Specifically, the U-Net model is applied for particle segmentation, achieving an intersection over union of 0.948 and enabling the extraction of morphology and the calculation of size. Then, a ResNet model is utilized for classification, achieving an overall accuracy of 96.3% across 12 types of particulate matter (including five MP types and seven non-MP particles), with the average accuracy for the five MP types reaching 97.2%. The critical role of polarization information is quantitatively validated through an ablation study, where its inclusion increases the classification accuracy by over 23% compared to using amplitude alone. The proposed method offers a rapid, reliable, and multi-parameter framework for the MP analysis, providing a powerful tool for environmental monitoring in aquatic systems.