Analysis of Wheat Spike Morphological Traits by 2D Imaging

Area of Science:

• Agricultural Science
• Plant Genetics
• Computational Biology

Background:

• Wheat spike morphology is crucial for grain yield, driving interest in genetics and breeding.
• Traditional measurement methods lack precision for complex spike phenotypes, hindering yield trait analysis.

Purpose of the Study:

• To develop a deep learning pipeline (Speakerphone) for precise acquisition of wheat spike phenotypes.
• To analyze the correlations between detailed spike morphology and wheat yield-related traits (thousand-grain weight and spike yield).

Main Methods:

• Developed and validated the Speakerphone deep learning pipeline for wheat spike segmentation (mIoU of 0.948).
• Validated phenotypic measurements against manual measurements (Pearson correlations > 0.96).
• Extracted 45 phenotypes from 221 wheat cultivars and performed hierarchical clustering.

Main Results:

• Speakerphone demonstrated high accuracy in spike segmentation and trait measurement, correlating strongly with manual data.
• Precise measurements of spike area and spikelet area clarified correlations between spike morphology and wheat yield.
• Hierarchical clustering identified six distinct wheat spike classes, revealing phenotypic variations affecting yield and thousand-grain weight.

Conclusions:

• Precise phenotyping of wheat spike morphology using deep learning enhances understanding of yield determinants.
• The study identified specific morphological traits influencing wheat yield and thousand-grain weight.
• Observed trends show an increase in large-spike cultivars over time, particularly in southern China, informing future breeding strategies.