Early diagnosis of wheat powdery mildew using solar-induced chlorophyll fluorescence and hyperspectral reflectance

Abstract

Powdery mildew disease threatens wheat production worldwide, and early detection is of great significance for disease control and maximizing yield and quality. To improve early remote sensing detection of wheat powdery mildew, solar-induced chlorophyll fluorescence (SIF) parameters were extracted using three-band Fraunhofer line discrimination (3FLD) and reflectance index approaches, and vegetation index (VI) was calculated by hyperspectral reflectance. All features and feature subsets of different data sources were used as inputs to multiple linear regression (MLR), random forest (RF), and support vector machine (SVM) algorithms to construct a wheat powdery mildew monitoring model. SVM includes linear kernel function (LK), polynomial kernel function (PK), and Gaussian radial basis function (RBF). Under wheat powdery mildew stress, wheat canopy reflectance showed a blue shift, and fluorescence weakened. The correlation between SIF−A intensity and disease index (DI) in the O²−A band extracted using the 3FLD method was the highest at −0.781, showing that the SIF parameter was useful for monitoring powdery mildew. Whether based on all features or feature subsets, the RBF model achieved the highest model accuracy, followed by the RF and the MLR. In the feature subset, the accuracy ranges of RBF, LK, and PK models are 0.740−0.871, 0.724−0.850, and 0.716−0.841 respectively. The SIF+VI in the RBF model is more useful for early and stable disease monitoring of wheat powdery mildew. This innovative technical solution is expected to support the early diagnosis of wheat powdery mildew, significantly improving disease prevention and control efficiency and effectiveness.