Identification of Fusarium head blight in wheat ears using vertical angle-based reflectance spectroscopy

Fusarium head blight (FHB) is a major disease that negatively affects wheat yield in China. Given that conventional reflectance spectroscopy measurements are perpendicular to crop canopy, the identification of FHB in wheat ears with the spectral data from the vertical angle can provide the possibility for large-scale monitoring. In this study, multi-features were selected and constructed to realize the identification of FHB in wheat ears from the vertical angle, and the influence of leafy and leafless samples were discussed. Firstly, the multi-features, such as band features, spectral position features, and vegetation indices for the leafy and leafless samples, were used to evaluate the ability to identify FHB, and correlation analysis was performed to select the effective features. In order to further reduce redundancy and enhance the separation capability of features, these candidate features were categorized into different feature sets based on Fisher score values. Then, the support vector machine (SVM) algorithm was used to construct the FHB identification model based on different feature sets of leafy and leafless samples. The optimal multi-features and the best classification accuracy were finally determined. The results were showed in the following: (1) The overall accuracies and Kappa coefficients of leafy samples could reach up to 65% and 0.28, respectively, whereas the values for the leafless samples could reach 81% and 0.63 in this model; (2) the optimal multi-features had great potential in identifying FHB-infected wheat ears; and (3) the presence of leaves would reduce the model’s identification capability and adversely affected the identification of FHB in wheat ears. These results provide realistic theoretical references for large-scale FHB monitoring, which are conducive to the selective harvest of wheat. © 2020, Saudi Society for Geosciences.