Deoxynivalenol (DON)
accumulation in wheat due to Fusarium Head Blight (FHB) negatively affects
grain quality and subsequently reduces grain yield. Evaluation of DON is an
integral part in breeding FHB-resistant wheat varieties. Here we explored the
potential of hyperspectral imaging to indirectly detect and predict DON
concentration. A total of 172 wheat genotypes evaluated for DON concentration
using GC/MS (Liquid Chromatography – Mass Spectrometry). Hyperspectral imaging
of Fusarium-damaged wheat kernels (FDKs) for each genotype was carried out
using a handheld hyperspectral imaging camera, Specim IQ (Specim Ltd., Oulo,
Finland). Processing of hyperspectral images was carried out using QGIS 3.10.2
and spectral reflectance values were carried out using Raster package in R. Of
the 204 wavebands (397 nm – 1004 nm) generated, genotypes showed significant
variation (p-value < 0.05) in 196 wavebands. However, only the first 130
wavebands (397 nm – 778 nm) were used for further analysis due to obvious noise
in the remaining wavebands. Pearson’s Correlation revealed significant
correlation (p-value < 0.05) between DON concentration and reflectance
values in all the 130 wavebands (r=0.32 to r=62). All 130 wavebands were used
in a simple Linear Regression Model generating an r2 value of 0.95. A
cross validation accuracy of actual vs. predicted value yielded an r and r2
value of 0.73 and 0.53, respectively. Similarly, Ridge Regression Best Linear
Unbiased Prediction (rrBLUP) yielded a cross validation accuracy of r=0.75. Five
wavebands: 622, 619, 628, 613, and 616 were identified through feature
selection to have the most contribution to observed variation. Taking the
results into account, this study has demonstrated the potential use of
hyperspectral imaging in detecting and predicting DON Concentration in
Fusarium-damaged wheat kernels.