VNIR Hyperspectral Signatures and Machine Learning for Early Detection and Classification of Barley Diseases.

Abstract

This study focuses on identifying barley diseases at various stages using the unique spectral signatures of phytopathogen infections. We examined the causal agents of widespread crop diseases, including: loose smut, head blight, fusarium head blight (FHB), stem rust, net blotch, spot blotch, common root rot. Analysing disease-specific spectral characteristics with machine learning (ML) algorithms revealed the most informative spectral ranges: the green region (~520-560 nm), the red chlorophyll absorption zone (~650-680 nm), and the red-edge region (~700 nm). These ranges accurately reflect alterations in the plant's cellular structure and pigment complexes. Spectral data were processed using five ML algorithms. Random Forest (RF) proved to be the most effective for identifying and differentiating barley diseases, achieving an accuracy of up to 90.13% (MCC = 0.86). This superior performance stems from the ensemble method's robustness to noise and its ability to extract critical features from high-dimensional hyperspectral data, particularly when distinguishing diseases with overlapping spectral signatures. Furthermore, this study highlights the potential of integrating UAV-based remote sensing to delineate reference zones, proximal hyperspectral imaging (HSI), and ML for robust plant health monitoring. This combined approach shows significant promise for early disease diagnostics, enabling site-specific treatments, curbing disease progression, and reducing pesticide application. Ultimately, these findings offer practical value for the agro-industrial sector in major grain-producing countries, especially in Central Asia, where agricultural advancement is a strategic priority for sustainable development and food security.

Plant phenotyping relevance

VNIRハイパースペクトル計測と機械学習により、オオムギの病害状態を植物体のスペクトル特徴から検出・分類する方法が研究の中心であり、植物フェノタイピング手法に該当する。

Code and data availability

The article describes hyperspectral barley disease data and ML analysis, but no public dataset, code, model, or image repository is provided. The Data Availability Statement says data are available only on request, and no author URLs or deposits appear in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.