An Automated Deep Learning Approach for Analyzing Stomatal Morphometry of Poplar Trees

In plants, stomatal pores regulate the intake of carbon dioxide that enables and powers photosynthesis and the loss of water from the surface of leaves. Variations in stomatal size and density can thus reflect novel strategies that balance water loss with photosynthetic capacity. Poplars are an emerging model system for studying adaptation to climate change in tree species and exhibit variation in stomatal characteristics. The use of deep learning algorithms such as the U-Net can aid in the reliable detection of key aspects of stomatal size and density in leaf tissue. The manual labor involved with studying plant stomata is a limiting factor due to the time it takes to analyze samples and inconsistency when taking measurements on the number of stomata, as well as their length, width, and area. Our approach is a "plug-and-play" method that takes a directory of images as an input and gives annotated images and measurements as an output. This is a simple-to-use method across disciplines. The framework allows for the modular replacement of trained deep learning models thereby enabling faster and more reproducible data collection in the future. Preliminary research is promising with meanIoU of 0.90 for training and 0.88 for validation data for capturing stomatal morphometry.