Related Experiment Video
Updated: Oct 16, 2025
11:37
RGB and Spectral Root Imaging for Plant Phenotyping and Physiological Research: Experimental Setup and Imaging Protocols
Published on: August 8, 2017
16.4K
Wheat physiology predictor: predicting physiological traits in wheat from hyperspectral reflectance measurements
Robert T Furbank1, Viridiana Silva-Perez2,3, John R Evans4
1ARC Centre of Excellence for Translational Photosynthesis, Research School of Biology. Australian National University, Canberra, ACT, 2601, Australia. robert.furbank@anu.edu.au.
Plant Methods
|October 20, 2021
Summary
New deep learning and ensemble models improve the prediction of wheat photosynthetic traits using hyperspectral data. These flexible models work across various spectral ranges, aiding crop breeding and yield prediction.
Area of Science:
- Plant science
- Spectroscopy
- Machine learning
Background:
- Crop breeding requires rapid in-field measurement of yield-contributing traits across many genotypes.
- Leaf hyperspectral reflectance data is used with partial least squares regression (PLSR) to predict genetic variation in wheat traits.
- Current PLSR models have limitations: fixed spectral input, trait-specific models, and reliance on expensive spectrometers.
Purpose of the Study:
- To compare the predictive accuracy of PLSR with deep learning and ensemble models.
- To assess model flexibility across different spectral ranges.
- To develop more accessible methods for predicting wheat physiological traits.
Main Methods:
- Trained and tested predictive models using previously published hyperspectral reflectance datasets.
- Compared partial least squares regression (PLSR) against various deep learning approaches and an ensemble model.
- Evaluated model performance and flexibility across different spectral wavelength ranges.
Main Results:
- Deep learning-based and ensemble models improved the accuracy of predicting photosynthetic and leaf traits in wheat compared to PLSR.
- These advanced models demonstrated flexibility, allowing application across different spectral ranges without significant accuracy loss.
- Overfitting was avoided with the developed deep learning and ensemble models.
Conclusions:
- The developed methods offer improved prediction of wheat leaf and photosynthetic traits from hyperspectral reflectance data.
- These approaches do not necessitate a full-range, high-cost leaf spectrometer, making them more accessible.
- A web service is provided for deploying these algorithms, supporting wheat yield prediction and crop breeding.
