High-throughput estimation of wheat tiller density using UAV-derived spectral-textural features and attention-based deep learning
Smart Agricultural Technology · 2026-12 · 15 · 102521
DOI: 10.1016/j.atech.2026.102521 ↗Kurz gefasst
Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Swas Kaushal und 13 weiteren Autor:innen. Die Arbeit nennt 77 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Sustainable Agriculture Research and Education, National Science Foundation, South Dakota State University.
Abstract (Originalsprache)
Wheat tiller density (TD) is a critical early-season indicator of crop growth and a key determinant of yield potential. However, current TD assessments still rely on manual counting, which is slow, costly, and difficult to scale. This study evaluates the use of Unmanned Aerial Vehicle (UAV)-acquired multispectral imagery combined with machine and deep learning-based predictive models for high-throughput estimation of wheat TD. A total of 721 TD observations were collected across five sites in South Dakota and paired with high-resolution UAV multispectral imagery. UAV-derived spectral and textural features were used to train a Random Forest regression (RFR) model and three deep learning models: Deep Neural Network (DNN), Convolutional Neural Network (CNN), and an attention-based CNN (Atten-CNN). Our results show that, on a held-out test set, Atten-CNN achieved the best performance (R² = 0.81; RMSE% = 15.97%), outperforming CNN (R² = 0.77; RMSE% = 17.74%), as well as RFR and DNN. Model interpretation using SHapley Additive exPlanations (SHAP) indicated that predictions were driven primarily by red-edge and pigment sensitive spectral features led by CRI_2 (Carotenoid Reflectance Index 2) and NDRE (Normalized Difference Red Edge Index), with additional contributions from GLCM (Gray Level Co-occurrence Matrix)-based canopy texture features, particularly band-wise correlation, variance, and entropy matrix. By integrating spectral and textural UAV features within an interpretable attention-based CNN and validating across breeding and commercial fields, this study demonstrates a practical, scalable framework for early-season wheat TD estimation.
Förderung
- Sustainable Agriculture Research and Education
- National Science Foundation
- South Dakota State University
Zitieren
APA
Kaushal, S., Maimaitijiang, M., Subedi, S., Thapa, S., Janjua, U. U. R., Koupal, D. J., Singh, M., Kaur, K., Kumar, P., Sitaula, P. R., Billah, M. M., Halder, J., Irshad, M. A., & Sehgal, S. K. (2026). High-throughput estimation of wheat tiller density using UAV-derived spectral-textural features and attention-based deep learning. Smart Agricultural Technology, 15, 102521. https://doi.org/10.1016/j.atech.2026.102521
Vancouver
Kaushal S, Maimaitijiang M, Subedi S, Thapa S, Janjua UUR, Koupal DJ, et al. High-throughput estimation of wheat tiller density using UAV-derived spectral-textural features and attention-based deep learning. Smart Agricultural Technology. 2026;15:102521. doi:10.1016/j.atech.2026.102521
BibTeX
@article{Kaushal2026,
author = {Kaushal, Swas and Maimaitijiang, Maitiniyazi and Subedi, Santosh and Thapa, Subash and Janjua, Ubaid ur Rehman and Koupal, Dante Jerome and Singh, Mandeep and Kaur, Karanjot and Kumar, Pradeep and Sitaula, Pradeep Raj and Billah, Mohammad Maruf and Halder, Jyotirmoy and Irshad, Muhammad Ali and Sehgal, Sunish Kumar},
title = {High-throughput estimation of wheat tiller density using UAV-derived spectral-textural features and attention-based deep learning},
journal = {Smart Agricultural Technology},
year = {2026},
volume = {15},
pages = {102521},
doi = {10.1016/j.atech.2026.102521}
}
EUVIMED