Abstract: Identification of wheat samples with six different levels of deoxynivalenol( DON) content by hyperspectral images,integrating stoichiometric method was studied in this paper. Hyperspectral images of 180 wheat samples were obtained,a Modified Gram- Schmidt algorithm( MGS) and a genetic uninformative variable elimination algorithm( GAUVE) were used to select sensitive wavelengths across the wavelength range of 400~1021 nm.Linear discriminant analysis( LDA),random forest( RF),support vector machine( SVM) and the K- nearest neighbors algorithm( KNN) models were established and developed to predict the DON content level of wheat samples. The results indicated that the MGS algorithm and GAUVE algorithm efficiently select the sensitive wavelengths,reduce the number of wavelength variables,and improve the operation rate. The accuracy rate of LDA algorithm,RF algorithm,SVM algorithm and KNN algorithm were found to be higher than 85%.Among all the identification models studied,MGS- SVM model obtained the best identification accuracy. This study research indicated that hyperspectral images combined with a stoichiometric method can accurately identify wheat kernels with six different levels of DON content,hence,offering a methodology for rapidly,non- destructively,intelligently detecting of wheat's DON toxin.