Automated Detection and Classification of Pollen Grains Using YOLO-Based Deep Learning Models

Area of Science:

• Palynology
• Computer Science
• Food Science

Background:

• Melissopalynology is crucial for honey authentication, verifying botanical origin, especially for high-value honey from regions like the Anzer Valley.
• Traditional manual pollen analysis is labor-intensive, time-consuming, and prone to observer bias, hindering efficient and standardized authentication processes.

Purpose of the Study:

• To develop and evaluate an automated system for simultaneous detection and classification of 53 ecologically and apiculturally important pollen types from the Anzer Valley.
• To compare the performance of two advanced deep learning object detection models, YOLOv11 and YOLOv12, for automated pollen grain identification.

Main Methods:

• A novel dataset of 2095 light microscopy pollen images was created and augmented to 19,860 images using geometric and photometric techniques.
• Two state-of-the-art object detection models, YOLOv11 and YOLOv12, were trained on the augmented dataset for pollen classification.
• Performance was evaluated using metrics such as mean Average Precision (mAP@0.5) and F1-score, with a focus on generalization to unseen data.

Main Results:

• The YOLOv11 architecture achieved superior performance over YOLOv12, with a mean Average Precision (mAP@0.5) of 93.2% and an F1-score of 0.896.
• The YOLOv11 model demonstrated high generalization capability on unseen data, achieving a recall of 90.6% and effectively handling complex microscopic sample visuals.
• Minimal misclassifications were observed, primarily between morphologically similar pollen grains within the Geraniaceae and Fabaceae families.

Conclusions:

• The developed YOLOv11-based system offers a reliable, high-throughput framework for automated pollen analysis, significantly reducing processing time and maintaining high accuracy.
• This automated approach provides a scalable tool for pollen identification, addressing the limitations of manual microscopy and supporting objective, standardized honey authentication and geographical origin verification.