Rapid Detection of Deoxynivalenol in Wheat Using 2D Raman Correlation Spectroscopy and Machine Learning

Fusarium Head Blight (FHB) poses a severe threat to wheat production and food safety in Kenya, leading to significant yield losses and contamination with the mycotoxin deoxynivalenol (DON).Current detection methods are either impractical for field use or lack the sensitivity to detect subsurface toxins.This study developed a rapid detection method by integrating conventional Raman spectroscopy with machine learning algorithms to identify DON-contaminated wheat kernels accurately.FHB symptomatic and healthy wheat heads were collected from farmers' fields, prepared via longitudinal sectioning to expose the endosperm, and analyzed using a portable 785 nm Raman spectrometer.Key spectral bands at 866, 1148, and 1456 cm⁻¹ were identified as DON biomarkers through two-dimensional correlation spectroscopy and statistical analysis.Principal Component Analysis (PCA) was employed for dimensionality reduction, with the first 16 principal components accounting for over 95% of the spectral variance.Supervised machine learning models, Support Vector Machine (SVM) and Random Forest (RF), were trained on PCA-transformed data to classify kernels as healthy or infected.The SVM model achieved 97% accuracy, while the RF model attained 100% accuracy on the test set.The entire workflow, from sample preparation to classification, required less than 8 minutes per kernel, demonstrating high throughput potential.The results indicate that conventional Raman spectroscopy, when combined with machine learning, shows potential as a rapid complementary screening approach for detecting DON contamination.With further validation, this method could support improved decision-making in wheat supply chains, particularly in resource-limited settings.