Related Experiment Video
Updated: Sep 27, 2025
08:11
Collection and Identification of Pollen from Honey Bee Colonies
Published on: January 19, 2021
7.5K
Detection and Recognition of Pollen Grains in Multilabel Microscopic Images
Elżbieta Kubera1, Agnieszka Kubik-Komar1, Paweł Kurasiński1
1Department of Applied Mathematics and Computer Science, University of Life Sciences in Lublin, Głęboka 28, 20-950 Lublin, Poland.
Sensors (Basel, Switzerland)
|April 12, 2022
Summary
This study introduces an automated method for pollen analysis using deep learning. The YOLO (You Only Look Once) network accurately identifies pollen grains in digital images, improving efficiency over manual methods.
Area of Science:
- Botany
- Computer Science
- Bioinformatics
Background:
- Manual pollen analysis is time-consuming and requires specialized expertise.
- Digital microscopy offers potential for automated analysis.
- Automating pollen identification can accelerate ecological and taxonomic research.
Purpose of the Study:
- To evaluate the effectiveness of automatic pollen analysis using deep learning.
- To compare the performance of YOLO with other object detection models.
- To assess the accuracy of automated pollen identification for key European taxa.
Main Methods:
- Digital microscopic images of pollen grains were analyzed using a deep neural network (YOLO).
- YOLO's recognition and detection capabilities were utilized, eliminating the need for image segmentation.
- Performance was benchmarked against Faster R-CNN and RetinaNet using mean average precision (mAP).
Main Results:
- YOLO achieved high accuracy, with mAP@.5:.95 ranging from 86.8% to 92.4%.
- YOLO outperformed Faster R-CNN and RetinaNet in pollen detection and classification.
- Challenges included similar grain morphology, co-occurrence, and overlapping of pollen grains.
Conclusions:
- Deep learning, specifically YOLO, provides an efficient and accurate automated solution for pollen analysis.
- Automated methods can overcome limitations of manual microscopic examination.
- Further development can address challenges like grain similarity and overlap for enhanced accuracy.
