Hyperspectral assessment of bacterial blight disease in red kidney beans by feature selection and machine learning algorithms

Abstract

PURPOSE: Bacterial blight poses a significant threat to red kidney bean growth, often leading to substantial yield losses. Real-time monitoring of this disease is crucial for effective prevention and control, ensuring optimal yield. This study proposes a hyperspectral-based approach to assess the the bacterial blight disease in red kidney beans. METHODS: In this study, canopy hyperspectral data from two experimental areas were collected, and five spectral preprocessing methods—multiple scattering correction (MSC), standard normal variation (SNV), first-order derivative (FD), second-order derivative (SD), and logarithmic transformation (LOG)—were applied to the raw spectra (R). Feature bands were extracted using a combination of the competitive adaptive reweighted sampling (CARS) and successive projection algorithm (SPA). Disease severity classification models were constructed using support vector machine (SVM) and partial least squares discriminant analysis (PLS-LDA), while estimation models were developed using support vector regression (SVR) and partial least squares regression (PLSR). RESULTS: Results demonstrated that FD preprocessing most effectively enhanced spectral features for bacterial blight detection, with sensitive bands optimally extracted using the CARS-SPA algorithm. The bands centered at 750 nm and 945 nm were identified as the most sensitive across all preprocessing methods. For condition index estimation, the PLSR model performed best, with FD preprocessing achieving R² values of 0.883 (modeling set) and 0.863 (validation set), and RMSE values of 0.072 and 0.083, respectively. CONCLUSIONS: These findings highlight the potential of hyperspectral technology, combined with feature extraction and machine learning algorithms, for efficient and accurate detection of bacterial blight in red kidney beans. This study provides a methodological and technical framework for monitoring other crops and diseases.

Plant phenotyping relevance

ハイパースペクトルデータからインゲンマメの病害状態・重症度を推定し、前処理、特徴帯抽出、分類・回帰モデルを検証することが研究の中心であるため、植物フェノタイピング手法として採用。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.