Quantitative phenotyping and evaluation for lettuce leaves of multiple semantic components

Area of Science:

• Plant Science
• Agricultural Engineering
• Computer Vision

Background:

• Accurate lettuce leaf phenotyping requires detailed quantification of multi-semantic traits, which is currently lacking.
• Existing manual methods are time-consuming, inaccurate, and fail to differentiate leaf components effectively.
• Automated, robust methods for extracting and validating lettuce leaf traits are needed for large-scale analysis.

Purpose of the Study:

• To develop an automated phenotyping pipeline for recognizing lettuce leaf components and calculating multi-semantic traits.
• To enable precise phenotype identification and comparative evaluation of lettuce leaf structures.
• To establish a foundation for high-throughput phenotyping applications in lettuce.

Main Methods:

• Developed six semantic segmentation models to extract leaf components from visible images.
• Implemented a leaf normalization technique for size-independent phenotyping.
• Utilized a lamina-based approach to identify petiole and vein structures (first and second order).

Main Results:

• The pipeline generated 30 geometry, 20 venation, and 216 color-based traits.
• Manually measured traits showed high correlation with computed results.
• Positive-back images validated the method's accuracy and evaluated trait differences.

Conclusions:

• The proposed method offers an effective strategy for quantitative analysis of detached lettuce leaf fine structure.
• Comprehensive utilization of geometry, color, and vein traits supports lettuce phenotype identification and breeding.
• This study advances automated high-throughput phenotyping for lettuce, improving traits like photosynthetic area and vein configuration.