Discrimination of winter wheat disease and insect stresses using continuous wavelet features extracted from foliar spectral measurements

Discrimination of crop diseases and insect damages is a critical task in pest management. As a non-contact and non-destructive method, spectroscopy has been recognised as an efficient way for crop pest detection. In this study, an advanced spectral analysis method, the continuous wavelet analysis (CWA), was used to discriminate three common diseases and insect damages in wheat crop: yellow rust, powdery mildew and aphid. In this research, leaf spectra were measured in both infected and reference plots at early grain filling stage. An algorithm was developed based on the continuously decomposed wavelet scalogram to identify types and severities of the damages. Its sensitivity and discrimination capability to damages were evaluated. Utilising an overlapping strategy, a wavelet feature selection method was established to identify optimal wavelet features discriminate the damages. Then, the discriminant model was developed based on the Fisher's linear discriminant analysis (FLDA). A total of six wavelet features with a central wavelength varying from 430 to 930 nm and scale factors of 4-8 were identified. According to a k-fold cross-validation, the averaged overall accuracy of the developed discriminant model was 77%. The CWA-based spectral discrimination approach showed good potential to serve as a basis to develop in-field, real-time, multi-damage mapping systems.