Automated multi-parameter characterization of microplastics using polarization digital holographic microscopy

Area of Science:

• Environmental Science
• Analytical Chemistry
• Optical Physics

Background:

• Microplastics (MPs) pose a significant threat to aquatic ecosystems globally.
• Current characterization methods for MPs lack speed and multi-parameter capabilities.

Purpose of the Study:

• To develop a novel, rapid, and multi-parameter technique for microplastic analysis.
• To automate the segmentation and classification of microplastics using deep learning.

Main Methods:

• Combined digital holography and polarization imaging to capture multi-dimensional particle features.
• Employed a deep learning pipeline (U-Net for segmentation, ResNet for classification) for automated analysis.
• Validated the importance of polarization information through an ablation study.

Main Results:

• Achieved high accuracy in particle segmentation (IoU 0.948) and classification (96.3% overall, 97.2% for MPs).
• Demonstrated that polarization information significantly improves classification accuracy by over 23% compared to amplitude alone.
• Successfully extracted morphology, size, and polarization features from single holographic images.

Conclusions:

• The proposed method provides a rapid, reliable, and multi-parameter framework for microplastic analysis.
• This technique offers a powerful tool for enhanced environmental monitoring of aquatic systems.
• The integration of optical imaging and deep learning advances microplastic identification capabilities.