Related Experiment Video
Updated: Jun 15, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
719
Automatic 3D Plant Organ Instance Segmentation Method Based on PointNeXt and Quickshift+
Sifan Dong1, Xueyan Fan1, Xiuhua Li1,2
1State Key Laboratory for Conservation and Utilization of Subtropical Agro-bioresources, School of Electrical Engineering, Guangxi University, Nanning 530004, China.
Plant Phenomics (Washington, D.C.)
|December 19, 2025
Summary
A novel two-stage method using PointNeXt and Quickshift++ achieves accurate organ instance segmentation for diverse plant types, advancing plant phenotyping research.
Area of Science:
- Computer Vision
- Plant Science
- Agricultural Technology
Background:
- Accurate organ instance segmentation of 3D plant point clouds is essential for plant phenotyping.
- Existing methods often lack generalization across different crop types (monocotyledonous vs. dicotyledonous).
Purpose of the Study:
- To develop a generalized two-stage method for single-plant organ instance segmentation.
- To improve the accuracy and applicability of plant organ segmentation across diverse species.
Main Methods:
- A two-stage approach combining an improved PointNeXt for semantic segmentation (stems, leaves) and Quickshift++ for instance segmentation.
- Training and validation on diverse datasets including sugarcane, maize, and tomato point clouds.
Main Results:
- Achieved high semantic segmentation accuracy (mOA 96.96%, mIoU 87.15%).
- Outperformed state-of-the-art methods in instance segmentation (mPrec 93.32%, mRec 85.60%, mF1 87.94%, mIoU 81.46%).
- Demonstrated strong generalization across different plant species and early growth stages.
Conclusions:
- The proposed method offers superior generalization for organ instance segmentation in 3D plant point clouds.
- This approach provides a robust tool for advancing plant phenotyping research across various crops.
