Detection of deoxynivalenol, its modified forms, and zearalenone in individual oat grains using visible-near-infrared spectroscopy and near-infrared hyperspectral imaging

Fusarium mycotoxins such as deoxynivalenol (DON), its modified forms, and zearalenone (ZEN) frequently contaminate oats, posing serious health and regulatory concerns. This study assessed the use of visible-near-infrared (Vis-NIR) spectroscopy and near-infrared hyperspectral imaging (NIR-HSI) to classify individual oat grains according to the European legal limits for DON (1,750 μg/kg) and ZEN (100 μg/kg). NIR-HSI consistently outperformed Vis-NIR, achieving classification accuracies (CAs) above 91% and F1-scores above 0.65 for DON, ZEN and combined DON+ZEN detection. The most informative spectral regions were in the NIR ranges of 1000-1250 nm and 1300-1500 nm, associated with Fusarium -induced biochemical and structural changes in oat grains. Reducing the spectral input to 20 selected wavelengths preserved NIR-HSI performance, supporting the feasibility of multispectral implementations. These targeted, non-destructive approaches could enable early removal of the few highly contaminated grains responsible for batch rejection, improving food safety, reducing waste, and enhancing the sustainability of oat processing. • Vis-NIR and NIR-HSI classified DON and ZEN-contaminated oats with >91% accuracy. • Removing ∼9% of grains could reduce DON-3G, 15-ADON and 3-ADON levels by >95%. • Optimal models relied on key wavelengths in the 1000-1200 and 1300-1500 nm ranges. • Reducing models to 20 wavelengths retained >92% accuracy across NIR-HSI methods. • Combined models enabled simultaneous detection of multiple Fusarium mycotoxins.