Determination and Visualization of Different Levels of Deoxynivalenol in Bulk Wheat Kernels by Hyperspectral Imaging

Abstract

A hyperspectral imaging system is proposed as a method to rapidly and nondestructively predict mycotoxin deoxynivalenol (DON) levels in FHB-infected wheat kernels. Standard normal variate transformation and multiplicative scatter correction (MSC) were used in spectral preprocessing. The successive projections algorithm (SPA) and random frog algorithm were used to select the optical wavelengths. Finally, the support vector machine (SVM) technique and partial least squares discriminant analysis were applied to establish different models for determining DON levels. Based on a comparison of the results, the MSC–SPA–SVM model, with the highest classification accuracy (100.00% for the training test and 97.92% for the testing set), gave the best performance, and a visualization map of the DON content level based on this model was created.