Optimizing the Experimental Method for Stomata-Profiling Automation of Soybean Leaves Based on Deep Learning

Area of Science:

• Plant Biology
• Microscopy
• Computational Biology

Background:

• Stomatal analysis is crucial for plant taxonomy, physiology, and ecology.
• Existing methods for stomatal imaging can be improved for clarity and automation.

Purpose of the Study:

• To develop an optimized leaf clearing method for enhanced soybean stomatal imaging.
• To create an automated stomatal detection system using deep learning.

Main Methods:

• Tested various ethanol and sodium hypochlorite (NaOCl) concentrations for leaf clearing.
• Selected absolute ethanol as a safer alternative fixative and dehydrating agent.
• Developed and trained a YOLO deep learning model for object detection of stomata.

Main Results:

• An optimal clearing formulation (1:1 mixture of 95% ethanol and 6-14% NaOCl) significantly improved soybean stomatal image quality.
• The YOLO model achieved high accuracy (mAP ~0.99) in automated stomatal detection.
• A web-based interface was created for accessible application of the detection model.

Conclusions:

• The new clearing method and deep learning approach provide an efficient and accurate tool for soybean stomatal analysis.
• This integrated system offers a valuable resource for plant science research, replacing toxic chemicals with safer alternatives.