An Affordable Image-Analysis Platform to Accelerate Stomatal Phenotyping During Microscopic Observation.

Abstract

Recent technical advances in the computer-vision domain have facilitated the development of various methods for achieving image-based quantification of stomata-related traits. However, the installation cost of such a system and the difficulties of operating it on-site have been hurdles for experimental biologists. Here, we present a platform that allows real-time stomata detection during microscopic observation. The proposed system consists of a deep neural network model-based stomata detector and an upright microscope connected to a USB camera and a graphics processing unit (GPU)-supported single-board computer. All the hardware components are commercially available at common electronic commerce stores at a reasonable price. Moreover, the machine-learning model is prepared based on freely available cloud services. This approach allows users to set up a phenotyping platform at low cost. As a proof of concept, we trained our model to detect dumbbell-shaped stomata from wheat leaf imprints. Using this platform, we collected a comprehensive range of stomatal phenotypes from wheat leaves. We confirmed notable differences in stomatal density ( SD ) between adaxial and abaxial surfaces and in stomatal size ( SS ) between wheat-related species of different ploidy. Utilizing such a platform is expected to accelerate research that involves all aspects of stomata phenotyping.

Code and data availability

The paper's stomata-detection GUI, trained SSD model weights, and model-training workflow are publicly available in the authors' GitHub repository (onsite_stomata_platform) with an executable Colab training notebook. The raw phenotype/image datasets are only available on request per the Data Availability Statement.