Multi-Source Data Fusion Improves Time-Series Phenotype Accuracy in Maize under a Field High-Throughput Phenotyping...

Area of Science:

• Agricultural Engineering
• Plant Science
• Remote Sensing

Background:

• High-throughput, 3D time-series phenotyping is vital for plant breeding and management.
• Challenges exist in aligning point cloud data and extracting accurate plant traits.

Purpose of the Study:

• To develop and validate a method for accurate, time-series 3D phenotyping of field maize populations.
• To assess the effectiveness of multi-source data fusion for improving phenotype extraction accuracy.

Main Methods:

• Utilized a rail-based phenotyping platform with LiDAR and RGB cameras for data acquisition.
• Aligned orthorectified images and LiDAR point clouds using direct linear transformation.
• Registered time-series point clouds with image guidance and removed ground points using the cloth simulation filter.
• Segmented individual plants and organs via fast displacement and region growth algorithms.

Main Results:

• Achieved high correlation (R² = 0.98) between fused multi-source data plant height measurements and manual measurements.
• Demonstrated improved accuracy compared to using single-source point cloud data (R² = 0.93).
• Validated the effectiveness of multi-source data fusion for enhancing time-series phenotype extraction.

Conclusions:

• Multi-source data fusion significantly improves the accuracy of time-series phenotype extraction in field conditions.
• Rail-based phenotyping platforms offer a practical solution for dynamic, individual plant and organ-scale phenotype observation.