A Novel Method for Filled/Unfilled Grain Classification Based on Structured Light Imaging and Improved PointNet+.
Shihao Huang1,2,3, Zhihao Lu1, Yuxuan Shi1
1College of Engineering, Huazhong Agricultural University, Wuhan 430070, China.
Sensors (Basel, Switzerland)
|July 29, 2023
Summary
This study introduces an improved deep learning method for classifying filled/unfilled rice grains using 3D point cloud data. The novel approach significantly enhances accuracy in rice grain identification for breeding and genetic analysis.
Area of Science:
- Agricultural Science
- Computer Vision
- Biotechnology
Background:
- Accurate classification of filled/unfilled rice grains is crucial for rice breeding and genetic analysis in China, the world's largest producer and consumer.
- Traditional manual methods for grain identification are inefficient, lack repeatability, and have low precision.
Purpose of the Study:
- To develop a novel, automated method for classifying filled/unfilled rice grains.
- To improve the efficiency and accuracy of rice grain analysis compared to traditional methods.
Main Methods:
- Acquired 3D point cloud data of rice grains using structured light imaging.
- Developed algorithms for single grain segmentation and normal vector-based data enhancement.
- Improved the PointNet++ deep learning network by adding a Set Abstraction layer and incorporating normal vector maximum pooling for classification.
Main Results:
- The Improved PointNet++ achieved a classification accuracy of 98.50%.
- This accuracy surpasses traditional machine learning models (e.g., XGboost at 91.99%) and other deep learning models like PointNet (93.75%) and PointConv (92.25%).
Conclusions:
- The study demonstrates a novel and highly effective method for filled/unfilled rice grain recognition using improved deep learning on 3D point cloud data.
- This automated approach offers significant advantages in precision and efficiency for agricultural applications.
