Early detection of rice blast disease using hyperspectral remote sensing
Plant Science Today · 19.01.2026 · 13 (1)
DOI: 10.14719/pst.11111 ↗Kurz gefasst
Fachartikel in der Zeitschrift Plant Science Today, online erschienen am 19.01.2026. Verfasst von R Selvaprakash und 4 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
We presented an integrative hyperspectral approach for the rapid and non-invasive detection of rice blast (Magnaporthe oryzae) that moves beyond traditional index-based methods. Leaf and canopy-level reflectance data (350 nm- 2500 nm) were smoothed using Savitzky - Golay polynomials, standardised with Standard Normal Variate (SNV) and Multiplicative Scatter Correction (MSC) and then differentiated to highlight subtle infection signals. Dimensionality reduction methods including Principal Component Analysis (PCA), t-Distributed Stochastic Neighbour Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) revealed clear separations between healthy and diseased spectra, while cosine similarity and the Spectral Angle Mapper (SAM) measured illumination-invariant spectral differences. A Random Forest impurity analysis identified the ten most informative wavelengths, enabling the evaluation of over one million band combinations. From this, we developed the Rice Blast Index (RBI = (R1068 - R1560) / (R1068 + R1560)), which outperformed Normalized Difference Vegetation Index (NDVI) and Photochemical Reflectance Index (PRI), achieving an F1-score of 0.95 and Cohen’s κ of 0.93 across independent growing seasons. New structural diagnostics, including lagged correlation, spectral autocorrelation and feature persistence, were introduced to quantify redundancy and identify stable biochemical absorption zones, notably a 38 nm region around 680 nm and a 1470 nm region linked with chlorophyll - protein features. Outlier spectra were removed with an Isolation Forest algorithm, improving robustness by 4.7 %. The average processing time was 18 ms per spectrum, enabling real-time scouting. Together, these elements deliver a unified, end-to-end framework that combines advanced pre-processing, dimensionality reduction, anomaly rejection, machine-learning-based band selection and new structural metrics. This framework improves early rice blast surveillance and offers a transferable template for hyperspectral phenotyping of diverse crop stresses, effectively bridging fine-scale sensitivity with field-scale applicability in precision agriculture.
Zitieren
APA
Selvaprakash, R., Rabi, N. S., Aundy, K., Rajeev, R., & Viswanathan, C. (2026). Early detection of rice blast disease using hyperspectral remote sensing. Plant Science Today, 13(1). https://doi.org/10.14719/pst.11111
Vancouver
Selvaprakash R, Rabi NS, Aundy K, Rajeev R, Viswanathan C. Early detection of rice blast disease using hyperspectral remote sensing. Plant Science Today. 2026;13(1). doi:10.14719/pst.11111
BibTeX
@article{Selvaprakash2026,
author = {Selvaprakash, R and Rabi, N S and Aundy, K and Rajeev, R and Viswanathan, C},
title = {Early detection of rice blast disease using hyperspectral remote sensing},
journal = {Plant Science Today},
year = {2026},
volume = {13},
number = {1},
doi = {10.14719/pst.11111}
}
EUVIMED