AUTHOR=Nagy Attila , Szabó Andrea , Elbeltagi Ahmed , Nxumalo Gift Siphiwe , Bódi Erika Budayné , Tamás János 

TITLE=Hyperspectral indices data fusion-based machine learning enhanced by MRMR algorithm for estimating maize chlorophyll content

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 15 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2024.1419316

DOI=10.3389/fpls.2024.1419316

ISSN=1664-462X

ABSTRACT=The research introduces an innovative method for accurate maize chlorophyll estimation, combining hyperspectral indices and machine learning models with the Minimum Redundancy Maximum Relevance (MRMR) algorithm. It tackles challenges in traditional chlorophyll estimation by utilizing rich hyperspectral data, aiming to monitor maize growth status and optimize nitrogen fertilizer applications in real time. The study explores a diverse set of hyperspectral indices, introducing new spectral indices covering various wavelengths to capture subtle variations in maize chlorophyll concentration. The proposed methodology employs six advanced machine learning models, including Robust Linear, Stepwise Regression, Support Vector Machines (SVMs), Fine Gaussian SVM, Matern 5/2 Gaussian Process Regression, and Trilayered Neural Networks, to exploit the complex relationships between hyperspectral indices and actual chlorophyll content. Integrating the MRMR algorithm enhances feature selection, optimizing the model's efficiency by identifying the most informative and least redundant spectral bands. The six machine learning models' performances were evaluated and compared using root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ). The Matern 5/2 Gaussian Process Regression model generated the highest prediction accuracy (training set: R 2 =0.71, RMSE=338.46 µg/g, andMAE=264.30 µg/g and validation set; R 2 =0.79, RMSE=296.37 µg/g, and MAE=237.12 µg/g). The results showcase that the Matern 5/2 Gaussian Process Regression model with MRMR can be used as the optimal model for estimating the maize chlorophyll and is a powerful tool for non-invasive and precise crop chlorophyll estimation. This research contributes to advancing precision detection of abiotic stress with proximal sensors, enabling timely and targeted interventions to optimize crop health and yield.