Near infrared hyperspectral imaging method to assess Fusarium Head Blight infection on winter wheat ears

Abstract

The scope of this article is to present a laboratory-based method for the assessment of Fusarium Head Blight (FHB) on ears of winter wheat using Near Infrared Hyperspectral Imaging (NIR-HSI) in the spectral range 900–1700 nm. Inoculated and non-inoculated winter wheat ears were collected from a trial field in Belgium and analysed in the laboratory. Using a dichotomous classification tree framework, Partial Least Squares Discriminant Analysis (PLS-DA) models were developed and applied at pixel level to discriminate the different parts of the ears, namely the awns, the stem and the healthy and infected parts of the ears. This study proposes the use of spectral information at pixel level instead of averaging the spectral signature of the ear. This approach allows using the spatial information of the data in order to provide a quantitative assessment of the severity of the infection on the ear. The results presented in this work confirm that wavelengths related to water and nitrogen are important for the detection of FHB. The validation of the models indicates a good performance for the detection of FHB infection at pixel level with a sensitivity of 96 % and a specificity of 100 %. Furthermore, the validation of the method at ear level shows good performance for the detection of FHB-infected ears with a sensitivity of 99.4 % and a specificity of 91.9 %. The literature reports similar performances, but using equipment with a broader spectral range (400–2500 nm). The present study thus indicates that good discrimination performance can be achieved using a much narrower spectral range. In addition, the obtained results suggest that the method can discriminate three classes of FHB severity, which enables a semi-quantitative assessment of the FHB infection.