Method for early diagnosis of verticillium wilt in cotton based on chlorophyll fluorescence and hyperspectral technology

Early and accurate detection of verticillium wilt (VW), the most common and devastating disease of cotton, is essential to prevent the spread of VW. However, it remains challenging to achieve accurate detection of VW in cotton before symptoms appear after infection with Verticillium dahliae (asymptomatic phase). This study evaluated the feasibility of detection of VW in the asymptomatic phase based on cotton main stem leaf chlorophyll fluorescence parameters (CFPs) and spectral features extracted based on continuous wavelet transform (CWT) in two different environments. The aim was to achieve accurate detection of cotton VW in the asymptomatic period by convenient methods. Hyperspectral data of cottons inoculated with V. dahliae were collected at different times, and the CFPs of main stem leaves were measured simultaneously. After preprocessing the hyperspectral data with CWT, common wavelet features for all spectral acquisition days and sensitive CFPs were extracted based on the results of ANOVA. Then, the variance inflation factor combined with least absolute shrinkage and selection operator (LASSO-VIF) was used to select the optimal wavelet features. Finally, the support vector machine, logistic regression, and k-nearest neighbors (KNN) were used to construct the models for detecting VW in asymptomatic leaves based on CFPs and optimal wavelet features, and the accuracy of the models were compared. The results showed that the CFPs were significantly affected 24 h after V. dahliae infection. V. dahliae infection reduced the maximum quantum yield (Pm') of photosystem II (PSII) and increased non-photochemical quenching (NPQt) in cotton leaves. Compared with the raw spectrum, the spectral features in the near-infrared region (800–1350 nm) extracted based on CWT could accurately reflect the subtle changes of leaves in the asymptomatic phase. Besides, compared with CFPs, the 4–5 wavelet features selected based on the LASSO-VIF were more helpful to accurately identify asymptomatic cotton leaves infected with V. dahliae, with an accuracy greater than 80 % and a Kappa coefficient higher than 0.6. Among them, the average accuracy of the logistic regression model based on wavelet features was as high as 90.62 %. The results of this study confirm the changes in CFPs in cotton leaves in the VW-asymptomatic period and the feasibility of accurate identification by using wavelet features. This study will provide a reliable reference for accurate large-scale identification of V. dahliae infection in cotton in the asymptomatic phase.