A new methodology of an automatic stomata classification and detection system

A Stomata Classification and Detection System in Microscope Images of Maize Cultivars

by Aono A. H., Nagai J. S., Dickel G. S., Marinho R. C., De Oliveira P. E. A. M., Faria F. A. (2019)

Alexandre H. Aono a, James S. Nagai a, Gabriella da S. M. Dickel b, Rafaela C. Marinho b, Paulo E. A. M. de Oliveira b, Fabio A. Faria a,∗

a Instituto de Ciéncia e Tecnologia, Universidade Federal de Sao Paulo – UNIFESP 12247-014, S˜ao José dos Campos, SP – Brazil

b Instituto de Biologia, Universidade Federal de Uberlandia
Uberlandia, MG, Brazil

===

In biorxiv – doi: http://dx.doi.org/10.1101/538165

https://www.biorxiv.org/content/biorxiv/early/2019/02/01/538165.full.pdf

Abstract

Stomata are morphological structures of plants that have been receiving constant attention. These pores are responsible for the interaction between the internal plant system and the environment, working on different processes such as photosynthesis process and transpiration
stream.

Figure 3: Examples of stoma (a) and non-stoma (b) subimages/regions, which were manually selected and
labeled in this work.

As evaluated before, understanding the pore mechanism play a key role to explore the evolution and behavior of plants. Although the study of stomata in dicots species of plants have advanced, there is little information about stomata of cereal grasses. In addition, automated detection of these structures have been presented on the literature, but some gaps are still uncovered.

Figure 4: In-depth explanation of the stomata identification process.

This fact is motivated by high morphological variation of stomata and the presence of noise from the image acquisition step.

Figure 5: Fifteen different microscope images of Maize Cultivars used in this work.

Herein, we propose a new methodology of an automatic stomata classification and detection system in microscope images for maize cultivars. In our experiments, we have achieved an approximated accuracy of 97.1% in the identification of stomata regions using classifiers based on deep learning features.

Figure 6: Different types of noise present in the microscopic images. (a) the usage of cyanoacrylate glue can
generate air bubbles; (b) leaves residuals might be captured by the microscope; (c) the leaves might bend and
generate grooves in the image; (d) degradated stomata due to biological factors; and (e) low image quality due
to equipment limitations.