Speaker
Description
Manual microscopic analyses are generally considered the gold standard for various palynological applications. However, in recent years, automated, database-supported methods for pollen analysis are increasingly being used. These approaches are considered to be more cost-effective, less time-consuming and enable better reproducibility than conventional microscope-based methods.
One example of such an innovative method is multispectral imaging flow cytometry (MIFC) combined with machine learning. MIFC is used to quickly record a large number of microscopic bright-field and fluorescence images, together with various pollen traits such as size, fluorescence and shape. This data can be used to train a convolutional neural network model to enable specific pollen identification.
In this study, the main pollen from 13 monofloral honeys (4x Tilia sp., 3x Brassica napus, 3x Castanea sativa, 1x Centaurea cyanus, 1x Helianthus annuus, 1x Phacelia tanacetifolia) was analysed using MIFC. A particular focus was placed on determining how pollen source (anthers vs honey), pollen traits (such as size, shape, fluorescence) and model input parameters influence classification performance. Various AI models were trained and tested for this purpose.
Clear differences between pollen sources were observed, especially in terms of size and fluorescence traits. These traits also proved to be particularly effective for distinguishing between pollen from different plant species. The best-performing MIFC model was compared to the results obtained through microscopy, revealing largely consistent classification outcomes with only minor deviations.
These findings suggest that MIFC is a promising tool for preliminary honey type screening and may serve as a complementary method to traditional microscopic pollen analysis. Future work will aim to adapt this approach for polyfloral honeys, where no dominant pollen type is present.
Keywords: pollen, honey, melissopalynology, multispectral imaging flow cytometry, AI
| Status Group | Doctoral Researcher |
|---|---|
| Poster Presentation Option | Undecided/No preference |
