Multi-dimensional machine learning approaches for fruit shape phenotyping in strawberry.

Area of Science:

• Agricultural Science
• Computer Science
• Genetics

Background:

• Strawberry fruit shape is crucial for visual appeal but is difficult to objectively measure.
• Current phenotyping relies on subjective human assessment, failing to capture the trait's complexity.
• Existing morphometric approaches are often abstract and hard to interpret for practical applications.

Purpose of the Study:

• To develop a mathematical approach for quantifying and classifying strawberry fruit shape.
• To transform complex shape data into human-recognizable categories using machine learning.
• To identify quantitative traits suitable for genetic analysis of fruit shape.

Main Methods:

• Utilized unsupervised machine learning to categorize strawberry fruit images into 4 principal shape groups.
• Applied the Principal Progression of k Clusters (PPKC) to infer an ordinal scale for shape progression.
• Extracted 68 quantitative features using morphometric analyses and multivariate statistics.

Main Results:

• Successfully transformed digital images into human-recognizable shape categories.
• Developed informative feature sets that accurately capture quantitative shape differences.
• Achieved high classification accuracy (68%–99%) for newly defined phenotypic variables.

Conclusions:

• Strawberry fruit shape can be robustly quantified, classified, and ordered using image analysis and machine learning.
• Generated a valuable dictionary of quantitative traits for genetic studies of fruit shape.
• The developed methods are applicable to other fruits, vegetables, and specialty crops.