Spectral response and intelligent classification of wheat leaves under powdery mildew stress

Objective  The response of wheat spectrum to powdery mildew stress based on hyperspectral technique was studied in order to determine the infection grade of powdery mildew quickly.

Method  The visible-near infrared spectra of wheat leaves were collected by fiber optic spectrometer combined with a integrating sphere and a leaf clip. The spectrum fitting SPAD (SF-SPAD) value was used to reflect the chlorophyll content, so as to preliminarily determine the infection of powdery mildew. Spectral sensitivity analysis was performed using PROSPECT model to identify sensitive bands. We combined dimension reduction by principal component analysis (PCA) and support vector machine (SVM) modeling to realize binary classification of spectral data. The infection degree of wheat was graded according to the percentage of disease spots determined by the PCA-SVM binary classification model.

Result  The SF-SPAD value increased with the increase of leaf order from bottom to top. Spots with SF-SPAD values less than 0.90 were disease spots, while spots with SF-SPAD values above 1.05 were good spots. The spectral sensitivity analysis identified the sensitive bands as 440−500 and 540−780 nm in the visible region, and therefore reduced the data dimension. The relationship between the infection grade (R) and the percentage of disease spots was determined as R₁: 0−30%,R₂: 30%−50%,R₃: 50%−70%,R₄: 70%−100%. The model established in this assay was suitable when the number of tested plants was above 20.

Conclusion  The monitoring model based on SF-SPAD and spectral PCA-SVM binary classification can accurately and rapidly determine the infection of wheat powdery mildew and the infection grade, reduce the number of samples, reduce the workload of detection on the ground, and improve the detection efficiency. The monitoring model is an intelligent monitoring technology which is practical, simple and easy to popularize.