Convolutional Neural Network and Hyperspectral Imaging for Detection of Deoxynivalenol Levels of Barley Kernels

Fusarium head blight (FHB) is a devastating disease impacting grain crops worldwide, leading to the production of mycotoxins in grains, with deoxynivalenol (DON) being the most prevalent. DON is present in both unprocessed and processed grains and poses a serious threat to human and animal health. Consequently, barley intended for wheat production, food, or feed must undergo routine testing for DON contamination. The study explored the potential of hyperspectral imaging (HSI) (wavelength range: 382–1030 nm) combined with convolutional neural networks (CNN) to classify DON concentrations in barley grains across various genetic lines. The proposed CNN model achieved 89.41% accuracy in distinguishing barley grains with low DON levels (<5 mg/kg) and high DON content (>5 mg/kg and ≤14 mg/kg). In addition, in the low DON category, 90.08% accuracy was achieved in the identification of Class I with DON ranging from 0.19 mg/kg to 1.25 mg/kg versus Class II DON ranging from 1.25 mg/kg to 5 mg/kg. These findings demonstrate that integrating HSI with CNN offers significant promise for differentiating DON levels in barley grains.