Abstract: Selecting the sensitive hyperspectral bands at different stages for monitoring of wheat stripe rust is beneficial to substantially improve the accuracy of disease monitoring. With one of the most important wheat diseases, the stripe rust, as an example, the present study proposed a framework of bands selection and optimization for disease detection at multiple stages. The study was conducted in a field experiment over three years. Based on the combination of independent t-test, correlation analysis, and the significance of component coefficients of Partial Least Square Discriminant Analysis (PLS-DA) and Partial Least Squares Regression (PLSR), the relative importance of bands for disease detection at specific stages were derived. For the efficient control of the stripe rust, the bands selection and optimization analysis was performed at two stages (i.e., early stage and middle stages) from the total three stages. Based on the identified bands at early and middle stages, a relatively satisfactory accuracy of 0.78 was achieved, with a root mean square error (RMSE) of 0.12. The results suggest that the hyperspectral technique has great potential in disease detection. The identified bands provide a basis for development of spectral index for disease detection in future.

Key words: wheat, stripe rust, hyperspectral, bands selection, multiple stages