Automatic Branch-Leaf Segmentation and Leaf Phenotypic Parameter Estimation of Pear Trees Based on Three-Dimensional...

Area of Science:

• Plant Science and Agricultural Technology
• Computer Vision and Machine Learning in Botany

Background:

• Leaf phenotypic traits are crucial for canopy photosynthesis efficiency.
• Traditional destructive sampling and manual measurements are time-consuming and hinder continuous plant growth monitoring.
• Non-destructive, accurate measurement of leaf traits is needed for efficient plant study.

Purpose of the Study:

• To propose an automatic branch-leaf segmentation pipeline using lidar point clouds.
• To conduct automatic measurements of leaf inclination angle, length, width, and area.
• To demonstrate the method's significance for monitoring pear tree growth and orchard management.

Main Methods:

• Established a 3D canopy model using lidar point cloud data via SCENE software.
• Employed the PointNet++ model for semantic segmentation of branches and leaves from point cloud data.
• Utilized a mean shift clustering algorithm for single leaf instance extraction and calculated leaf parameters via plane fitting and triangulation.

Main Results:

• Achieved high accuracy in semantic segmentation (mean IoU of 0.88) and leaf instance extraction (mCoV of 0.87).
• Demonstrated strong correlations between automatically measured and manually measured leaf parameters (correlation coefficients ranging from 0.91 to 0.94).
• Quantified leaf inclination, length, width, and area with high precision, showing minimal root mean squared errors.

Conclusions:

• The proposed lidar-based pipeline enables automatic and accurate measurement of pear leaf phenotypic parameters.
• This non-destructive approach significantly aids in continuous monitoring of plant growth and simulation of canopy photosynthesis.
• The method offers substantial benefits for optimizing orchard management and agricultural practices.