Early diagnosis of Cladosporium fulvum (syn. Fulvia fulva) in tomato using the visible-near-infrared hyperspectral imaging

Plant diseases, such as Cladosporium fulvum (C. fulvum) disease, severely affect crop growth and yield, and early diagnosis of disease is more important than disease diagnosis to maintain crop yield and quality. In this study, C. fulvum was artificially inoculated, and hyperspectral images of healthy and infected samples were obtained using a visible-near-infrared (VIS/NIR) hyperspectral imaging system. Spectral data from the regions of interest (ROIs) were extracted and pre-processed, and competitive adaptive reweighted sampling (CARS), bootstrapping soft shrinkage (BOSS), and CARS-BOSS algorithms were used for feature extraction. The performance of recognition models was built using random forest (RF), extreme learning machine (ELM), and partial least squares discriminant analysis (PLS-DA) classification models, which were compared using the data before and after feature extraction. The results indicated that the superior performance of the ELM model, which achieved accuracies of 99.78% and 95.39% for calibration and prediction sets, with respective Kappa values of 0.99 and 0.98. After CARS-BOSS feature extraction, the model achieved accuracies of 98.68% and 99.34% for calibration and prediction sets, respectively, with a Kappa value of 0.99 for both. This study demonstrated that combining VIS/NIR hyperspectral technology with the CARS-BOSS-ELM model provided a rapid nondestructive method for early diagnosis of C. fulvum in the tomato.