[Quantitative Identification of Yellow Rust and Powdery Mildew in Winter Wheat Based on Wavelet Feature].

Area of Science:

• Agricultural Remote Sensing
• Plant Pathology
• Spectroscopy

Background:

• Powdery mildew and stripe rust cause significant winter wheat yield losses globally.
• Accurate quantitative disease identification is crucial for effective fungicide application.
• Current methods may lack the precision for differentiating specific diseases.

Purpose of the Study:

• To investigate the potential of hyperspectral data and continuous wavelet transform for quantitative distinction of powdery mildew and yellow rust.
• To develop and validate models for differentiating these diseases at the canopy level.
• To provide a remote sensing-based approach for crop disease identification.

Main Methods:

• Hyperspectral data acquisition and spectral normalization.
• Continuous wavelet transform (CWT) for feature extraction (spectral bands and wavelet features).
• Correlation analysis, t-tests, principal component analysis (PCA), and Fisher linear discriminant models.
• Model validation using leave-one-out and 5-5 sample cross-validation.

Main Results:

• Continuous wavelet features (WFs) significantly improved classification accuracy (90.4%-92.7%) compared to spectral bands (SBs) alone (61.5%-65.5%).
• Models combining SBs and WFs achieved the highest overall accuracies (91.1%-94.6%).
• Yellow rust was discriminated with up to 100% user's and producer's accuracy using WFs and SB&WFs models.

Conclusions:

• Continuous wavelet features show great potential for discriminating different plant disease stresses.
• Hyperspectral imaging combined with CWT offers a robust method for quantitative crop disease identification.
• This approach provides a theoretical basis for wide-range crop disease monitoring using remote sensing.