Yellow rust (Puccinia striiformis f. sp. Tritici) and powdery mildew (Blumeria graminis) are two serious diseases that severely intimidated yield and grain quality of winter wheat around the world. Since the preferable habitat conditions of them are similar, there is a high possibility that both diseases occurred in field simultaneously. To facilitate a differentiation of control procedures (i.e. using different fungicide), the discrimination of yellow rust and powdery mildew is a necessity. As a fast and nondestructive technique in obtaining the plant status information in real time, remote sensing has been successfully applied in the monitoring of crop diseases in several cases. However, studies addressing the discrimination of different crop diseases are rare. The aim of the present work was to assess the capability of remote sensing in discriminating yellow rust and powdery mildew at leaf level. For each disease, a total of 30 leaf samples were collected for spectral measurement, including both infected and non-infected leaves. Prior to the analysis, the spectral data were undertaken a normalization, to minimize the spectral difference caused by the cultivars. The spectra of normal leaves were compared with that of both infected ones (yellow rust and powdery mildew) through an independent t-test. Within the bands that were significantly different between normal and diseased leaves, a further band selection was conducted to differentiate powdery mildew from yellow rust using the same independent t-test. Only those disease sensitive bands that have the discriminative power were retained. Their discriminative capability was examined by a fisher linear discrimination analysis (FLDA). It turned out that the discrimination model yielded satisfactory estimation of sample categories, with an overall accuracy over 0.9. Therefore, it is evident that the hyperspectral remote sensing is a promising way to discriminate yellow rust and powdery mildew.
展开▼
