The monitoring and identification of tomato early blight during the incubation period by hyperspectral image

Early detection of tomato early blight during its incubation period is essential for precision agriculture and effective disease management. This study integrates hyperspectral imaging and machine learning techniques to diagnose early symptoms of tomato early blight, focusing on improving spectral preprocessing, feature selection, and model robustness. Hyperspectral images of tomato leaves inoculated with Alternaria solani were captured, and the earliest identifiable time during the incubation period was determined to be 28 hours post-inoculation using Euclidean distance analysis. Various spectral preprocessing methods were compared, and the combined SG-MSC-STD method was selected to optimize signal-to-noise ratios. A novel feature wavelength selection method was proposed by combining the Grey Wolf Optimizer (GWO) with machine learning models. Successive Projections Algorithm (SPA), Competitive Adaptive Reweighting Algorithm (CARS), and GWO were tested for feature selection. Four classifiers—Partial Least Squares Discriminant Analysis (PLS-DA), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Light Gradient Boosting Machine (LightGBM)—were evaluated using both full-spectrum and feature wavelengths. The GWO-LightGBM model achieved the highest accuracy of 97.22%, and its robustness was confirmed by cross-validation and ROC curve analysis. Furthermore, the physiological relevance of selected feature wavelengths was analyzed to understand the spectral mechanisms linked to plant disease progression. This study demonstrates the feasibility of early detection of tomato early blight during the incubation period using hyperspectral imaging and machine learning. The proposed GWO-based feature selection method and optimized spectral preprocessing significantly enhanced model accuracy and interpretability, offering valuable insights into the physiological basis of disease development.