Fast and Efficient Root Phenotyping via Pose Estimation.

Area of Science:

• Plant biology
• Computer vision
• Bioinformatics

Background:

• Image segmentation is standard for plant phenotyping but is labor-intensive and error-prone.
• Existing methods require extensive data annotation for training segmentation models.
• Geometric features derived from segmentation masks are sensitive to mask accuracy.

Purpose of the Study:

• To develop a segmentation-free approach for plant root phenotyping using deep learning.
• To automate the detection of morphological landmarks on plant roots.
• To enable accurate and efficient root trait extraction and analysis.

Main Methods:

• Leveraged deep learning-based landmark detection and grouping (pose estimation) using SLEAP (Social LEAP Estimates Animal Poses).
• Applied the method to plant roots using a gel cylinder imaging system across multiple species.
• Developed the Python library 'sleap-roots' for trait extraction from pose data.

Main Results:

• The pose estimation approach reliably and efficiently recovered root system topology with high accuracy.
• Achieved high accuracy using fewer annotated samples and at a faster speed compared to segmentation-based methods.
• Pose-derived root traits demonstrated high accuracy and utility in downstream tasks like genotype classification.

Conclusions:

• Established the validity and advantages of pose estimation for plant phenotyping, offering a more efficient and accurate alternative to segmentation.
• The 'sleap-roots' library facilitates direct comparison of pose-derived traits with segmentation-based analyses.
• The developed tools, data, and code are publicly available to encourage adoption and further research.