DFSP: A fast and automatic distance field-based stem-leaf segmentation pipeline for point cloud of maize shoot

Area of Science:

• Agricultural Engineering
• Computer Vision
• Plant Science

Background:

• 3D point cloud data analysis is crucial for plant morphology and phenotypic estimation.
• Automated, high-precision, and fast plant point cloud segmentation remains a challenge.
• Integrating global and local features efficiently in point cloud analysis is difficult.

Purpose of the Study:

• To develop a rapid and accurate method for plant organ segmentation using 3D point cloud data.
• To enable precise stem-leaf segmentation for maize phenotype research.
• To create a pipeline that integrates global spatial structure and local morphological features.

Main Methods:

• Developed a distance field-based segmentation pipeline (DFSP) for plant point cloud analysis.
• Extracted terminal point clouds for organ segmentation and identified maize stem base using local geometric features.
• Utilized regional growth and DFSP for stem point cloud generation and leaf instance segmentation.

Main Results:

• DFSP achieved an average processing time of 1.52 seconds for approximately 15,000 maize plant points.
• The segmentation algorithm demonstrated high performance with mean precision (0.905), recall (0.899), and micro F1 score (0.902).
• The method was validated on 420 maize samples, showing accurate and reliable stem-leaf segmentation.

Conclusions:

• DFSP provides an accurate, rapid, and automated solution for maize stem-leaf segmentation from 3D point clouds.
• The pipeline effectively integrates global and local features for robust organ segmentation.
• DFSP shows significant potential for advancing automated phenotyping in maize research.