Evaluating the potential of airborne hyperspectral imagery in monitoring common beans with common bacterial blight at different infection stages

Abstract

Common bacterial blight (CBB) is the most destructive bacterial disease affecting the production of common beans, and timely detection of CBB is crucial to limiting its spread. In this study, correlation analysis and the ReliefF algorithm were used to select vegetation indices (VIs) and texture features (TFs) that are sensitive to CBB. The CBB monitoring model based on support vector machine regression (SVR), random forest regression (RFR), and K-nearest neighbor regression (KNNR) was established using the selected the VIs, TFs, and their combinations. Then, the impact of the spatial resolution on the disease monitoring accuracy was evaluated. In addition, the early infection monitoring model was further optimised. The results show that in the early infection stage, when the spatial resolution was 0.07 m, the window size was 7 × 7, and the independent variable was a combination of VIs and TFs, the R² of the monitoring model constructed via SVR was 0.72, which was 14.3% higher than that obtained for a 3 × 3 window (0.63). In the middle and late infection stages, the optimal spatial resolution was 0.1 m, and the monitoring model constructed using RFR and a combination of VIs and TFs performed the best, with R² values of 0.81 and 0.88, respectively. The research results indicate that selecting an appropriate spatial resolution and window size can effectively improve the model's CBB monitoring ability and can provide a reference for accurate monitoring of large-scale CBB of common beans using airborne or spaceborne imaging spectroscopy technology.