Abstract:The study sought to establish a method for the rapid identification of mold contamination in wheat using 126 wheat samples.Near infrared spectroscopy combined with stoichiometry was used to establish a discriminant model based on support vector machine（SVM）classification.The SVM method was based on the elimination of abnormal samples，spectral reduction，and pretreatment.Five abnormal samples were eliminated by principal component analysis based on Mahalanobis distance.Eight principal components were obtained by reducing the dimension of the original spectral data，which represented 98.80%of the original samples.The best preprocessed method of input variables was standard normal variable transformation.The best kernel function was linear with a kernel function parameter C value of 10.The accuracy rate of training set discrimination of the SVM discriminant model was 100%and the cross-verification discrimination accuracy was 98.89%.The external verification set samples were used to verify the SVM discriminant model.The discrimination accuracy of the SVM discriminant model was 100%for the external verification set samples.The SVM discriminant model established in the study could be used for the rapid detection of mold contamination in wheat.