Continuous wavelet analysis based spectral features selection for winter wheat yellow rust detection

Remote-sensing technologies can provide quick responses for determining the presence of yellow rust disease. However, most studies selected spectral indicators solely based on the statistical relevance to disease severity. Few of them investigated the underlying mechanism including the variation of biochemical properties due to the presence of disease. This study aims at identifying some mechanism based spectral features through continuous wavelet (CWT) analysis. An inoculation of yellow rust fungal was conducted in the experimental field. The hyperspectral data and biochemical properties including the content of water and pigment were measured for both infected and non-infected winter wheat plants. A two-tailed paired student t-test was used to identify the biochemical properties which have significant variation when the plants were infected. For those sensitive biochemical properties, a CWT transformation was then processed with the spectral data. The sensitive spectral features were selected through a correlation scalogram. It was found that the content of chlorophyll decreased significantly in yellow rust infected plants. Four spectral features were identified which could well reflect the chlorophyll content. The predicted model of chlorophyll content was thus established based on the partial least squares (PLS) regression. In addition, the linear discrimination analysis (LDA) was adopted in classifying the infected and non-infected plants. The classification accuracy reached 74.8% which indicated the selected spectral features have great potential in detecting the winter wheat yellow rust infection.