Related Experiment Video
Updated: Nov 9, 2025
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.6K
Deep Segmentation of Point Clouds of Wheat
Morteza Ghahremani1,2, Kevin Williams1, Fiona M K Corke1
1National Plant Phenomics Centre, Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Aberystwyth, United Kingdom.
Frontiers in Plant Science
|April 12, 2021
Summary
A new deep neural network, Pattern-Net, effectively segments 3D wheat point clouds into organs for direct trait analysis. This groundbreaking method enables detailed plant structure modeling in three dimensions.
Area of Science:
- Agricultural Science
- Computer Vision
- Computational Biology
Background:
- 3D plant analysis is crucial for modeling organ structure and traits.
- Segmenting unstructured 3D point clouds presents significant challenges due to the lack of regular grids.
Purpose of the Study:
- To introduce Pattern-Net, a novel pattern-based deep neural network for segmenting 3D point clouds of wheat.
- To enable the direct analysis of wheat organ traits within 3D space.
Main Methods:
- Pattern-Net utilizes a K-nearest neighbor algorithm to establish dynamic links between neighboring points.
- The network creates multi-level abstraction patterns, connecting layers to enhance link propagation and mitigate the vanishing-gradient problem.
- It is designed to decompose complex, unstructured point clouds into semantically meaningful components.
Main Results:
- The study demonstrates the first successful segmentation of wheat point clouds into defined organs.
- Pattern-Net effectively analyzes and decomposes unstructured 3D point cloud data.
- Experiments confirm the approach's efficacy in 3D wheat segmentation.
Conclusions:
- Pattern-Net offers a robust solution for segmenting complex 3D plant point clouds.
- This method facilitates direct, in-situ trait analysis of wheat organs in 3D space.
- The approach advances 3D plant modeling and analysis capabilities.
