Hyperspectral imaging combined with machine learning for high‐throughput phenotyping in winter wheat
Abstract Uncrewed aerial vehicles (UAVs) are a highly successful and efficient method for phenotyping in plant breeding programs. This study explored using UAVs equipped with hyperspectral sensors to expedite breeders' decision‐making in selecting winter wheat (Triticum aestivum L.) genotypes w...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2024-12-01
|
| Series: | Plant Phenome Journal |
| Online Access: | https://doi.org/10.1002/ppj2.20111 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Uncrewed aerial vehicles (UAVs) are a highly successful and efficient method for phenotyping in plant breeding programs. This study explored using UAVs equipped with hyperspectral sensors to expedite breeders' decision‐making in selecting winter wheat (Triticum aestivum L.) genotypes with improved growth, biomass, leaf area, and canopy cover (CC). The hyperspectral image processing pipeline utilized for image analysis was described. The study was conducted with 2145 genotypes of winter wheat, and UAV‐based hyperspectral measurements were used to predict the phenotype. Reflectance measurements were taken at narrow wavelength intervals, spanning 400–2500 nm. For ground truthing, samples were collected from different parts of the field. These samples were used to evaluate various plant attributes, including CC, leaf area index, plant height, and dry biomass. The hyperspectral data were employed for computation of multiple vegetation indices (VIs), and to improve the prediction of plant traits, we employed partial least squares regression (PLSR) and random forest (RF) regression techniques on both the complete set of hyperspectral variables and the top 10 derived VIs. Our results show that using complete hyperspectral variables results in superior r, R2, and lower root mean square error for both models. We conclude that relying solely on linear regression models with VIs may not always result in accurate predictions of plant traits in winter wheat. However, combining these indices with the RF and PLSR algorithm significantly enhances the prediction accuracy. However, the performance of both PLSR and RF models shows minimal disparity, with only slight differences observed. This highlights the importance of utilizing appropriate machine learning algorithms for improved prediction of plant traits. |
|---|---|
| ISSN: | 2578-2703 |