Related Experiment Video
Updated: Jul 1, 2026
12:06
Analyzing Mitochondrial Morphology Through Simulation Supervised Learning
Published on: March 3, 2023
Deep learning-based 3D morphological segmentation and quantitative growth analysis of field-grown cabbage across the
Hongda Li1,2, Chaoguo Si2,3, Yisheng Miao2,3,4
1College of Information and Electrical Engineering, China Agricultural University, Beijing, 100083, China.
Plant Phenomics (Washington, D.C.)
|June 30, 2026
Summary
This study introduces a novel 3D point cloud segmentation network for precise, full-cycle monitoring of field-grown cabbage. The method accurately captures plant growth dynamics, advancing precision agriculture and digital twin applications.
Area of Science:
- Agricultural Science
- Computer Vision
- Robotics
Background:
- Conventional 2D monitoring lacks spatial data for precise, full-cycle cabbage phenotyping in open fields.
- Limitations in current methods hinder accurate growth dynamics tracking and precision agriculture.
- There is a need for advanced techniques to capture 3D structural characteristics of crops.
Purpose of the Study:
- To develop a high-precision 3D point cloud dataset for monitoring cabbage growth from seedling to maturity.
- To propose an adaptive point cloud segmentation network for whole-cycle growth monitoring.
- To enable accurate extraction of phenotypic traits for dynamic growth analysis.
Main Methods:
- Constructed a 3D point cloud dataset using depth cameras and multi-view spatial registration.
- Developed an adaptive point cloud segmentation network with Head Refinement Module (HRM), Leaf Instance Segmentation Module (LISM), and Cross Module Interaction (CMI).
- Validated the method on field-grown cabbage, evaluating semantic and instance segmentation performance and growth parameter extraction.
Main Results:
- The proposed network achieved high accuracy in semantic segmentation (mIoU 0.767, accuracy 94.8%) and instance segmentation (AP improvement for heads and leaves).
- Extracted growth parameters (plant height, canopy spread) showed strong correlation (R² > 0.9) with ground-truth measurements.
- The method demonstrated superior performance over state-of-the-art models in segmentation tasks.
Conclusions:
- The study successfully enables dynamic, full-cycle monitoring of cabbage phenotypes using 3D point cloud data in open-field conditions.
- This research provides a novel technical pathway for extending 3D phenotyping from controlled environments to practical agricultural applications.
- The findings support precise crop monitoring and the development of digital twin agriculture systems.