StomaVision: stomatal trait analysis through deep learning

Abstract

Summary StomaVision is an automated tool designed for high-throughput detection and measurement of stomatal traits, such as stomatal number, pore size, and closure rate. It provides insights into plant responses to environmental cues, streamlining the analysis of micrographs from field-grown plants across various species, including monocots and dicots. Enhanced by a novel collection method that utilizes video recording, StomaVision increases the number of captured images for robust statistical analysis. Accessible via an intuitive web interface at and available for local use in a containerized environment at , this tool ensures long-term usability by minimizing the impact of software updates and maintaining functionality with minimal setup requirements. The application of StomaVision has provided significant physiological insights, such as variations in stomatal density, opening rates, and total pore area under heat stress. These traits correlate with critical physiological processes, including gas exchange, carbon assimilation, and water use efficiency, demonstrating the tool’s utility in advancing our understanding of plant physiology. The ability of StomaVision to identify differences in responses to varying durations of heat treatment highlights its value in plant science research. Plain language summary StomaVision is a tool that automatically counts and measures tiny openings on plant leaves, helping us learn how plants deal with their surroundings. It is easy to use and works well with various plant species. This tool helps scientists see how plants change under stress, making plant research easier and more accurate.

Plant phenotyping relevance

気孔数、孔サイズ、閉鎖率などの植物形質を画像から自動抽出するツールの開発・提供が研究の中心であり、植物フェノタイピング手法に該当する。

Code and data availability

The authors publicly release their StomaVision source code, trained YOLOv7-seg model, and all labeled stomata images on GitHub, plus a public Streamlit web portal for stomatal trait analysis. Cited datasets (Dryad/LeafNet, Cuticle Database) and generic libraries (VDP, Detectron2, Ultralytics, Label Studio) are prior/th