Automated Phenotypic Trait Extraction for Rice Plant Using Terrestrial Laser Scanning Data.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• Accurate measurement of rice plant phenotypic traits is crucial for crop breeding and management.
• Traditional methods for phenotyping are often labor-intensive and time-consuming.
• Terrestrial Laser Scanning (TLS) offers a non-destructive, high-throughput approach for capturing 3D plant data.

Purpose of the Study:

• To develop an automated computational process for extracting six key rice plant phenotypic traits from TLS data.
• To propose an effective method for extracting the tiller number in rice plants.
• To design and implement a user-friendly, automated phenotype extraction tool for rice plants.

Main Methods:

• Utilized terrestrial laser scanning (TLS) data for 3D point cloud acquisition of rice plants.
• Developed a three-layer architecture software tool using PyQt5 and Open3D for automated phenotype extraction.
• Implemented algorithms to compute crown diameter, stem perimeter, plant height, surface area, volume, and projected leaf area.
• Developed a specific method for tiller number extraction.

Main Results:

• High reliability was observed for four verified features, indicated by strong linear coefficients of determination (R²).
• Root Mean Square Error (RMSE) for crown diameter, stem perimeter, and plant height was at the centimeter level.
• RMSE for tiller number was as low as 1.63.
• Relative RMSE (RRMSE) for crown diameter, plant height, and tiller number was within 10%, with stem perimeter at 18.29%.

Conclusions:

• The developed automated tool efficiently extracts phenotypic features from rice plant point clouds.
• The study provides a convenient and rapid method for obtaining critical rice plant traits.
• Future research will focus on expanding sample data and refining accuracy algorithms for broader applicability.