LenRuler: a rice-centric method for automated radicle length measurement with multicrop validation

Area of Science:

• Agricultural Science
• Computer Vision
• Biotechnology

Background:

• Radicle length is crucial for predicting seed vigor, germination, and seedling growth.
• Current measurement techniques lack automation, efficiency, and broad applicability across diverse seed types.

Purpose of the Study:

• To develop an automated, accurate, and generalizable method for radicle length measurement.
• To address the limitations of manual and inefficient existing measurement approaches.

Main Methods:

• Utilized the Segment Anything Model (SAM) with a coarse-to-fine segmentation strategy and Gaussian-based classification.
• Automated bounding box and centroid generation for precise seed coat and radicle segmentation.
• Calculated radicle length using geodesic distance between skeletonized seed coat and radicle.

Main Results:

• Achieved high segmentation performance on rice seeds (Dice: 0.955, Pixel Accuracy: 0.944).
• Demonstrated high precision in rice radicle length measurement (MAE: 0.273 mm, R²: 0.982).
• Validated cross-species performance in maize, pearl millet, and rye.

Conclusions:

• LenRuler provides an accurate and automated solution for rice radicle length measurement.
• The method shows significant potential for application in other crop species.
• Automated radicle length analysis supports enhanced seed quality assessment and crop improvement.