Label3DMaize: toolkit for 3D point cloud data annotation of maize shoots.

Area of Science:

• Plant science
• Computer vision
• Computational biology

Background:

• 3D point clouds are crucial for plant structure and morphology studies.
• Accurate plant organ segmentation from point clouds is essential for phenotype estimation and 3D reconstruction.
• Current automatic and robust point cloud segmentation methods for plants are lacking, hindering high-throughput analysis.

Purpose of the Study:

• To develop a novel point cloud segmentation algorithm for maize shoots.
• To create a user-friendly toolkit for semi-automatic point cloud annotation and segmentation of maize shoots.
• To improve the accuracy and efficiency of plant organ-level phenotyping.

Main Methods:

• A top-to-down point cloud segmentation algorithm utilizing optimal transportation distance was developed.
• The Label3DMaize toolkit was created for semi-automatic segmentation and annotation of maize shoots.
• The process involves stem segmentation, coarse segmentation, fine segmentation, and sample-based segmentation.

Main Results:

• Label3DMaize enables semi-automatic segmentation and annotation of maize shoots across different growth stages.
• Segmentation of a single maize shoot takes approximately 4-10 minutes, with coarse segmentation requiring 10-20% of total time.
• Coarse segmentation accuracy reached 97.2%, with fine segmentation offering greater detail, especially at organ connections.

Conclusions:

• Label3DMaize integrates segmentation algorithms and manual operations for semi-automatic maize shoot point cloud processing.
• The toolkit serves as a practical data annotation tool, facilitating deep learning-based segmentation research.
• This approach is expected to advance automatic point cloud processing for diverse plant species.