New Optimized Spectral Indices for Identifying and Monitoring Winter Wheat Diseases

The vegetation indices from hyperspectral data have been shown to be effective for indirect monitoring of plant diseases. However, a limitation of these indices is that they cannot distinguish different diseases on crops. We aimed to develop new spectral indices (NSIs) that would be useful for identifying different diseases on crops. Three different pests (powdery mildew, yellow rust, and aphids) in winter wheat were used in this study. The new optimized spectral indices were derived from a weighted combination of a single band and a normalized wavelength difference of two bands. The most and least relevant wavelengths for different diseases were first extracted from leaf spectral data using the RELIEF-F algorithm. Reflectance of a single band extracted from the most relevant wavelengths and the normalized wavelength difference from all possible combinations of the most and least relevant wavelengths were used to form the optimized spectral indices. The classification accuracies of these new indices for healthy leaves and leaves infected with powdery mildew, yellow rust, and aphids were 86.5%, 85.2%, 91.6%, and 93.5%, respectively. We also applied these NSIs for nonimaging canopy data of winter wheat, and the classification results of different diseases were promising. For the leaf scale, the powdery mildew-index (PMI) correlated well with the disease index (DI), supporting the use of the PMI to invert the severity of powdery mildew. For the canopy scale, the detection of the severity of yellow rust using the yellow rust-index (YRI) showed a high coefficient of determination ( ${\mbi{R}}^{\bf 2 =} {\bf 0.86}$ ) between the estimated DI and its observations, suggesting that the NSIs may improve disease detection in precision agriculture application.