Related Experiment Video
Updated: Jul 19, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.5K
Extraction of soybean plant trait parameters based on SfM-MVS algorithm combined with GRNN
Wei He1, Zhihao Ye2, Mingshuang Li3
1College of Engineering, Nanjing Agricultural University, Nanjing, China.
Frontiers in Plant Science
|August 10, 2023
Summary
This study introduces a 3D reconstruction method for soybean plants using smartphone images and structure from motion (SFM). The technique accurately measures plant traits and aids in variety identification, offering a faster, more objective alternative to manual methods.
Area of Science:
- Agricultural Science
- Computer Vision
- Plant Phenomics
Background:
- Soybean is a globally significant crop, vital for nutrition and industry.
- Accurate phenotypic measurement is crucial for soybean breeding and yield improvement.
- Current manual measurement methods are subjective, labor-intensive, and slow.
Purpose of the Study:
- To develop a non-destructive, automated 3D reconstruction and phenotyping method for soybean plants.
- To overcome the limitations of manual phenotypic measurements in soybean research.
- To enable accurate measurement of morphological traits and identification of soybean varieties.
Main Methods:
- Utilized structure from motion (SFM) to reconstruct 3D point clouds of soybean plants from smartphone images.
- Applied data fusion techniques (low-pass filtering, Gaussian filtering, OLS, Laplacian smoothing) for automatic segmentation of plant structures (stems, leaves).
- Developed a leaf phenotype measurement (LPM) algorithm for quantifying eleven morphological traits and employed machine learning models (SVM, BP, GRNN) for variety prediction.
Main Results:
- The 3D reconstruction method achieved high accuracy in measuring plant height, leaf length, and leaf width, with low root mean square error (RMSE) and mean absolute percentage error (MAPE).
- Coefficients of determination (R2) for measured traits were high, indicating strong correlation with actual values.
- The GRNN model achieved the highest accuracy (0.9211) in predicting soybean plant species based on leaf parameters.
Conclusions:
- The proposed 3D reconstruction and phenotyping method effectively extracts detailed phenotypic information from soybean plants non-destructively.
- This automated approach offers significant improvements in speed, objectivity, and accuracy compared to manual measurements.
- The method shows potential for application in phenotyping other dense-leaved plants, advancing high-throughput plant research.
