Spectral analysis of winter wheat leaves for detection and differentiation of diseases and insects

Authors

Lin Yuan, Beijing Academy of Agriculture and Forestry Sciences
Yanbo Huang, USDA-ARS
Rebecca W. Loraamm, University of South Florida
Chenwei Nie, Beijing Academy of Agriculture and Forestry Sciences
Jihua Wang, Beijing Academy of Agriculture and Forestry Sciences
Jingcheng Zhang, Beijing Academy of Agriculture and Forestry SciencesFollow

Date of this Version

2014

Citation

Field Crops Research 156 (2014) 199–207

Comments

© 2013 Elsevier B.V.

This document is a U.S. government work and is not subject to copyright in the United States.

http://dx.doi.org/10.1016/j.fcr.2013.11.012

Abstract

Yellow rust (Puccinia striiformis f. sp. Tritici), powdery mildew (Blumeria graminis) and wheat aphid (Sitobion avenae F.) infestation are three serious conditions that have a severe impact on yield and grain quality of winter wheat worldwide. Discrimination among these three stressors is of practical importance, given that specific procedures (i.e. adoption of fungicide and insecticide) are needed to treat different diseases and insects. This study examines the potential of hyperspectral sensor systems in discriminating these three stressors at leaf level. Reflectance spectra of leaves infected with yellow rust, powdery mildew and aphids were measured at the early grain filling stage. Normalization was performed prior to spectral analysis on all three groups of samples for removing differences in the spectral baseline among different cultivars. To obtain appropriate bands and spectral features (SFs) for stressor discrimination and damage intensity estimation, a correlation analysis and an independent t-test were used jointly. Based on the most efficient bands/SFs, models for discriminating stressors and estimating stressor intensity were established by Fisher’s linear discriminant analysis (FLDA) and partial least square regression (PLSR), respectively. The results showed that the performance of the discrimination model was satisfactory in general, with an overall accuracy of 0.75. However, the discrimination model produced varied classification accuracies among different types of diseases and insects. The regression model produced reasonable estimates of stress intensity, with an R² of 0.73 and a RMSE of 0.148. This study illustrates the potential use of hyperspectral information in discriminating yellow rust, powdery mildew and wheat aphid infestation in winter wheat. In practice, it is important to extend the discriminative analysis from leaf level to canopy level.