A Two-Stage, Semi-Supervised Deep Learning Framework for the Detection and Classification of Ambient Pollen using...

Area of Science:

• Environmental Science
• Biotechnology
• Microscopy

Background:

• Airborne pollen, a significant aeroallergen, poses public health risks, exacerbated by climate change-induced shifts in pollen dynamics.
• Accurate monitoring of airborne pollen is crucial for effective public health interventions and allergy management.
• Current methods like bright-field microscopy offer limited detail for precise pollen identification.

Purpose of the Study:

• To introduce and evaluate evanescent wave scattering microscopy (EWSM) for label-free, high-resolution identification of airborne pollen.
• To develop and implement a two-stage semi-supervised framework for autonomous pollen identification and classification.
• To refine the classification system through a human-in-the-loop approach for improved accuracy and scalability.

Main Methods:

• Utilized evanescent wave scattering microscopy (EWSM) to capture unique scattering signatures of pollen particles, visualizing fine morphological details.
• Implemented a two-stage semi-supervised learning framework employing RT-DETR for pollen localization and EfficientNetB0 for classification.
• Applied a "human-in-the-loop" strategy for iterative label refinement, enhancing model performance across two phases.

Main Results:

• The RT-DETR model achieved an 82.3% recall rate for pollen localization.
• Phase 1 refinement improved the F1 score from 0.277 to 0.537 for 41 species, reaching 80% recall.
• Phase 2 consolidation into 19 morphological groups yielded a precision of 0.648, recall of 0.744, and F1-score of 0.693, reducing false positives by 63%.

Conclusions:

• EWSM provides detailed morphological insights for label-free airborne pollen identification, surpassing traditional microscopy.
• The developed semi-supervised framework with human-in-the-loop refinement offers a scalable and accurate solution for autonomous pollen analysis.
• This approach holds significant potential for real-time environmental monitoring and public health strategies in urban settings.