Abstract

In this paper, we address the problem of preparing data for training detectors to identify transparent objects in light microscopy images. To this end, we propose using blends of reference images and monitoring background, instead of time-consuming labelling of monitoring data. This approach allowed us to avoid the need to involve a palynologist in the preparation of the training data while also ensuring 100% correct ground-truth labels. The statistical analysis of the deep learning results confirms that the results obtained for blends only are more stable, and in some cases surpass the results obtained for the training set with some labelled monitoring data added to reference images and monitoring background.

Recommended Citation

Kubera, E., Wieczorkowska, A., Kubik-Komar, A., Piotrowska-Weryszko, K. & Konarska, A. (2024). Image Processing for Improving Detection of Pollen Grains in Light Microscopy Images. In B. Marcinkowski, A. Przybylek, A. Jarzębowicz, N. Iivari, E. Insfran, M. Lang, H. Linger, & C. Schneider (Eds.), Harnessing Opportunities: Reshaping ISD in the post-COVID-19 and Generative AI Era (ISD2024 Proceedings). Gdańsk, Poland: University of Gdańsk. ISBN: 978-83-972632-0-8. https://doi.org/10.62036/ISD.2024.61

Paper Type

Full Paper

DOI

10.62036/ISD.2024.61

Download

Share

COinS
 

Image Processing for Improving Detection of Pollen Grains in Light Microscopy Images

In this paper, we address the problem of preparing data for training detectors to identify transparent objects in light microscopy images. To this end, we propose using blends of reference images and monitoring background, instead of time-consuming labelling of monitoring data. This approach allowed us to avoid the need to involve a palynologist in the preparation of the training data while also ensuring 100% correct ground-truth labels. The statistical analysis of the deep learning results confirms that the results obtained for blends only are more stable, and in some cases surpass the results obtained for the training set with some labelled monitoring data added to reference images and monitoring background.