Discrimination of Pests and Diseases in Winter Wheat Based on Pigment Sensitive Indices

Abstract:

In 2020, the global food security has been greatly threatened due to the coronavirus disease 2019 (COVID-19), climate change and crop diseases and pests (CDP). Among them, the occurrence of CDP is closely related to global climate change, and the COVID-19 will also have an indirect impact on the occurrence of CDP. Climate warming and the high moisture may cause the various changes of some fungal pathogens, e.g. yellow rust (Puccinia striiformis) and powdery mildew (Blumeria graminis). And also, aphid is one of major wheat pests which has a wide range of distribution and a large area of occurrence. A series of changes of pigment content will occur when winter wheat is disturbed by CDP. In this paper, the difference in the content of chlorophyll (Chl), carotenoids (Car) and carotenoid/chlorophyll (Car/Chl) under the different CDP stresses was explored; and the pigment sensitive indices were used to assess the abilities of the discrimination of the health wheat and wheat infected by yellow rust (YR), powdery mildew (PM) and aphid (AH). As the result showed, both of the Chl content of PM and YR were higher than healthy wheat, stripe rust's Car content was slightly lower than healthy wheat. The Car/Chl of the three kinds of stress is higher than that of healthy vegetation. Because when vegetation is under stress, due to the protection mechanism, Car/Chl has a certain increase. In the discrimination part of the research, support vector machine (SVM), k-Nearest Neighbor (KNN)and linear discriminate (LDA) are selected for model construction. The classification accuracy of chlorophyll sensitive indices is the highest, reaching 97.3%, followed by Car/Chl ratio. The SVM classification accuracy of pigment indices was 93.5%-97.3%, which was all higher than the common stress indices. Among the three classifiers, SVM achieves the best classification accuracy in the classification of four groups of indices. The results of Car/Chl sensitive indices under all three classifiers were all good, and different classifiers had less influence on the classification results than other indices. The classification effect is more stable. In the analysis of confusion matrix, Chl sensitive indices can distinguish healthy samples and stripe rust samples with high accuracy, and the Car sensitive indices had the highest accuracy in identifying stripe rust.