Feasibility of applying hyperspectral imaging as a rapid method for estimating potassium content across different plant species datasets

This is the authors' abstract. We don't add key points for this paper.

Purpose The accurate and frequent estimation of the leaf plant potassium concentration enabled by hyperspectral imaging techniques has allowed growers to optimize fertilizer applications and reduce the negative impact on the environment. In this study, we examined the feasibility of using leaf spectral data to accurately estimate the potassium content in sugar beet and celery plants. Methods Leaf images in the visible and near infrared region (VNIR: 400–1000 nm) and short-wavelength infrared region (SWIR: 1000-2500 nm) were captured by a hyperspectral camera. The potassium content was measured by ordinary destructive laboratory methods. The correlation-based feature selection (CFS) algorithm was implemented to select important wavelengths that carried the most useful information for predicting the potassium content in plant leaves. Four multivariate regression methods were tested to find a model with strong predictive performance. Results The experimental results showed that the Random Forest (RF) model using 12 bands (425, 443, 479, 599, 631, 662, 798, 863, 897, 921, 1978 and 2053 nm) had the highest accuracy for predicting potassium content in sugar beet, celery and both plant datasets (Rp2 = 0.85, Rp2 = 0.79, and Rp2 = 0.81, respectively). Conclusion The results confirm the universality of the described method. Although further validation studies involving other plant species are needed, it appears that the spectral reflectance technique could be a promising tool for the rapid, noninvasive and cost-effective estimation of K content in plant leaves, contributing to a significant step forward in precision fertilization management.