Using high-throughput phenotype platform MVS-Pheno to reconstruct the 3D morphological structure of wheat

Area of Science:

• Agricultural Science
• Plant Science
• Computer Vision

Background:

• Understanding plant morphology is crucial for crop yield and resource management.
• Three-dimensional (3D) information offers a more accurate representation of plant structures than 2D methods.
• Automated 3D data acquisition is a key challenge in plant phenotyping.

Purpose of the Study:

• To develop a point cloud data-driven 3D reconstruction method for wheat plants.
• To achieve phytomer-scale 3D structure reconstruction and morphology parameterization.
• To enhance the accuracy and efficiency of 3D plant analysis for agricultural applications.

Main Methods:

• Utilized the MVS-Pheno platform for wheat plant point cloud reconstruction.
• Employed deep learning algorithms for organ segmentation (semantic and instance).
• Automated 3D reconstruction of leaves and tillers, followed by morphological parameter extraction.

Main Results:

• Achieved 95.2% semantic segmentation accuracy for plant organs.
• Obtained an instance segmentation accuracy (AP50) of 0.665.
• High R² values (0.80-1.00) for extracted morphological parameters like leaf length, width, tiller length, and spike length.

Conclusions:

• The proposed method significantly improves the accuracy and efficiency of 3D wheat plant morphological analysis.
• Provides robust technical support for agricultural production optimization and genetic breeding.
• Enables detailed, phytomer-level phenotyping for advanced crop research.