Abstract: Based on the hyperspectral imaging technology, the non-destructive and rapid identification method for maize seed purity was proposed. The data were preprocessed by multiple scattering correction（MSC）, and the competitive adaptive reweighted sampling（CARS） and iteratively retains informative variables（IRIV） were used to extract the characteristic wavelengths. The purity identification models of support vector machine（SVM） and line discriminant analysis（LDA） were established. The random seed value was set, and the points to be evaluated in the confidence interval were searched by the acquisition function of expected-improvement-plus to obtain the hyper parameter value with the minimum cross-validation loss for improving the model accuracy. The results show that the MSC-IRIV-LDA recognition model has the highest accuracy. The accuracies of the training set and the prediction set are respective 0.960 4 and 0.933 3, and the Kappa coefficient is 0.918 6. After optimizing the Delta and Gamma hyper parameters of LDA, the accuracies of training set and prediction set and Kappa coefficient can be further improved. The proposed method can realize non-destructive and rapid identification of maize seed purity, which can provide technical support for the development of precision agriculture.