Reconstruction of hyperspectral information from low-cost UAV RGB images to improve alfalfa yield prediction

Authors: LANG, QIAO - University Of Wisconsin; JIAHAO, FAN - University Of Wisconsin; Franco Jr, Jose; Duff, Alison; Diaz Vallejo, Emily; ZHOU, ZHANG - University Of Wisconsin

Submitted to: Biosystems Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/31/2025
Publication Date: 11/18/2025
Citation: Lang, Q., Jiahao, F., Franco Jr, J.G., Duff, A., Diaz Vallejo, E.J., Zhou, Z. 2025. Reconstruction of hyperspectral information from low-cost UAV RGB images to improve alfalfa yield prediction. Biosystems Engineering. https://doi.org/10.1016/j.biosystemseng.2025.104328.
DOI: https://doi.org/10.1016/j.biosystemseng.2025.104328

Interpretive Summary:

Technical Abstract: Alfalfa, as an important high-quality feed around the world, requires timely and accurate yield prediction for precision agricultural management. Unmanned aerial vehicle (UAV) hyperspectral remote sensing (RS) provides feasible solutions for obtaining alfalfa yield due to its advantages of repeated and high throughput observations. However, data redundancy, high costs and poor robustness of the monitoring models are still major obstacles to the widespread application of hyperspectral RS. To overcome these challenges, this study proposes a hyperspectral sensitive information reconstruction method based on RGB images to achieve low-cost, high-precision alfalfa yield prediction. In the 2021 experiments, hyperspectral and RGB sensors carried by UAV were used to collect RS images of the alfalfa canopy, and a total of 180 yield samples were collected. Firstly, three different methods were used to optimize sensitive bands for alfalfa yield. Secondly, hyperspectral sensitive bands were reconstructed based on RGB images and deep learning models, and the reconstruction performance was evaluated. Finally, alfalfa texture features were extracted from RGB images and fused with reconstructed hyperspectral sensitive features to construct a high-precision alfalfa yield model. To further determine the universality and transferability of the proposed method, field samples collected in 2022 were utilized as target task to validate the proposed method. Validation results demonstrated that the method proposed in this study could effectively predict alfalfa yield with good accuracy and robustness, providing a feasible approach for low cost, high precision crop monitoring based on UAV RS.