RotatedStomataNet: a deep rotated object detection network for directional stomata phenotype analysis.

Area of Science:

• Plant science
• Computational biology
• Image analysis

Background:

• Stomata are crucial for plant gas exchange and development.
• Current methods for stomatal analysis are often inaccurate or inefficient, especially for dense or unevenly distributed stomata.
• High-throughput, intelligent stomatal detection is essential for plant research.

Purpose of the Study:

• To develop an end-to-end, real-time system for intelligent stomatal detection and identification.
• To address limitations of existing methods in accurately analyzing stomatal density and aperture size.
• To enable automated, high-throughput phenotyping of stomata.

Main Methods:

• Developed RotatedeStomataNet, treating stomatal detection as a rotated object detection problem.
• Utilized destructive and non-destructive datasets from Arabidopsis and maize.
• Implemented an automated system for collecting stomatal and aperture phenotypic data.

Main Results:

• RotatedeStomataNet provides accurate, real-time detection and identification of stomatal density and aperture size.
• The system demonstrated high accuracy across monocotyledon (maize, wheat) and dicotyledon (Arabidopsis, soybean) species.
• Experimental results showed consistency between automated predictions and manual labeling.

Conclusions:

• RotatedeStomataNet offers an innovative, efficient solution for stomatal phenotyping.
• The system enables one-stop automatic collection of phenotypic data without complex operations.
• This technology advances plant research by providing accurate and high-throughput stomatal analysis.