Early Detection and Classification of Phytophthora Blight in Chili Peppers Using Hyperspectral Imaging and Machine Learning

Phytophthora blight, caused by Phytophthora capsici, is difficult to control after symptoms appear, motivating rapid, non-destructive detection during the asymptomatic stage. This study evaluated hyperspectral imaging (HSI) for early disease detection in chili pepper seedlings. Stem and leaf spectra were monitored during disease progression, and the basal stem was selected for its disease-related spectral changes before visible symptoms appeared. A genetic algorithm coupled with partial least squares and correlation analysis identified five characteristic wavelengths (550, 670, 722, 760, and 800 nm). Among the six machine-learning algorithms evaluated on day 5 using five-fold cross-validation, linear discriminant analysis (LDA) and support vector machine (SVM) achieved classification accuracies of 90.7% and 90.1%, respectively. Detection before symptom onset was feasible by day 5. Healthy and asymptomatic infected samples collected on that day were therefore evaluated using 20 vegetation indices to identify simpler diagnostic predictors. Single-index support vector machine (SVM) prediction accuracies ranged from 60.00% to 91.25%. The photochemical reflectance index (PRI) achieved the highest accuracy (91.25%), followed by the enhanced vegetation index (EVI) and triangular vegetation index (TVI) (90.00% each). These findings support basal-stem HSI and selected vegetation indices as candidate approaches for early P. capsici detection, subject to independent validation.