Deep Detection Models for Measuring Epidermal Bladder Cells

Epidermal bladder cells (EBC) are specialized structures of halophyte plants that accumulate salt and other metabolites, and are thought to be involved in salinity tolerance, as well as in UV-B protection, drought and other stresses tolerance. However, the role of the EBC size, density or volume remains to be confirmed since few studies have addressed the relevance of these traits. This is due to the fact that those measurements are mostly carried out manually. In this work, we have tackled this problem by conducting a statistical analysis of several deep learning algorithms to detect EBC on images. From such a study, we have obtained a model, trained with the YOLOv4 algorithm, that achieves a Fl-score of 91.80%. Moreover, we have proved the reliability of this model to other varieties and organs using different datasets than the ones used for training the model. In order to facilitate the use of the YOLO model, we have developed LabelGlandula, an open-source and simple-to-use graphical user interface that employs the YOLO model. The tool presented in this work will help to understand the functioning of EBC and the molecular mechanisms of EBC formation and salt accumulation, and to transfer this knowledge to crops one day.