Automatic pre-screening of outdoor airborne microplastics in micrographs using deep learning.

Area of Science:

• Environmental Science
• Data Science
• Toxicology

Background:

• Airborne microplastics (AMPs) are widespread in indoor and outdoor settings, presenting potential human health risks.
• Automated identification of AMPs in micrographs is crucial for research and monitoring.
• Existing deep learning studies on microplastics often use high-resolution images from aquatic environments.

Purpose of the Study:

• To develop and evaluate deep learning models for identifying and classifying outdoor AMPs in low-resolution micrographs.
• To integrate classification within segmentation frameworks for improved efficiency.
• To offer a faster, more accurate alternative to traditional AMPs identification methods.

Main Methods:

• Employed enhanced U-Net models (Attention U-Net, Dynamic RU-NEXT) and Mask Region Convolutional Neural Network (Mask R-CNN).
• Focused on low-resolution micrographs (256 × 256 pixels) of outdoor AMPs.
• Integrated classification directly into U-Net-based segmentation for streamlined analysis.

Main Results:

• Enhanced U-Net models achieved >85% classification F1-score and >77% segmentation accuracy.
• Mask R-CNN yielded 73.32% bounding box precision, 84.29% classification F1-score, and 71.31% mask precision.
• The proposed deep learning methods outperformed traditional thresholding techniques.

Conclusions:

• The study presents an efficient and accurate deep learning approach for AMPs identification in low-resolution images.
• This method serves as an effective pre-screening tool, reducing the need for extensive chemical analysis.
• The findings advance the monitoring and characterization of airborne microplastics through AI integration.