ABSTRACT
Stomata fulfill an important physiological role and are often phenotyped by researchers in many fields. Currently, no fully automated method exists to perform this task. Researchers typically rely on manual counts of stomata, which is an error-prone method and difficult to reproduce.
We introduce StomataCounter, an automated stomata counting system using a deep convolutional neural network to identify pores in a variety of different microscopic images. We used a human-in-the-loop approach to train and refine a neural network on a large variety of microscopic images, which helps us achieve robust detection among a number of datasets.
Our network achieves 98.1% identification accuracy on Ginkgo SEM micrographs, and 94.2% transfer accuracy when tested on untrained species.
To facilitate adoption of the method, we make a web tool available under http://www.stomata.science/
