A Deep Learning Method for Fully Automatic Stomatal Morphometry and Maximal Conductance Estimation.

Area of Science:

• Plant physiology and anatomy
• Computational biology and machine learning

Background:

• Stomata are crucial for plant gas exchange, influencing overall plant performance.
• Maximal stomatal conductance (gsmax) is a key functional property determined by stomatal anatomy (size and density).
• Manual measurement of stomatal dimensions is laborious, time-consuming, and prone to errors.

Purpose of the Study:

• To develop and validate an automated method for estimating stomatal morphometry from leaf impressions.
• To predict the functional property of anatomical maximal stomatal conductance (gsmax) using automated measurements.
• To establish a rapid and repeatable pipeline for plant phenotyping and gas flux assessment.

Main Methods:

• Development of a deep learning network utilizing semantic segmentation for stomatal morphometry.
• Creation of an automated pipeline integrating image analysis and trait measurement.
• Validation using leaf impression datasets from wheat and poplar species.

Main Results:

• The deep learning model achieved 100% accuracy in distinguishing and detecting stomata in wheat and poplar datasets.
• Automated gsmax estimations closely matched expert manual calculations, with a 3.8% difference for wheat and 1.9% for poplar.
• Semantic segmentation proved to be a rapid and repeatable method for anatomical gsmax estimation.

Conclusions:

• Automated morphometry of stomata from leaf impressions via deep learning offers a significant advancement over manual methods.
• This approach effectively estimates anatomical gsmax, addressing a key bottleneck in plant phenotyping.
• The developed pipeline enables rapid assessment of plant gas exchange based on stomatal morphology.