Due to the lack of confirmation technology for rapid identification of rice origin, 186 rice samples from Wuchang, Northeast and South of China, were identified by near-infrared mid-infrared, and Raman combined with chemometric analysis in this study. Firstly, the recognition effects of three algorithms of k-nearest neighbor method (KNN), linear discriminant analysis (LDA) and least squares-support vector machine (LS-SVM), combined with five preprocessing methods on three single spectrum rice origin identification models are compared. The results show that the Raman spectrum model of the LS-SVM algorithm combined with the SNV+2nd preprocessing method is the best, and the accuracy of the calibration set and validation set are 100% and 93.48% respectively. In order to further improve the accuracy of the identification model, the rice-origin identification models of data layer fusion, feature layer fusion and decision-layer fusion based on near-infrared spectroscopy, mid-infrared spectroscopy and Raman spectroscopy are established innovatively. The results show that the recognition accuracy of the three spectral information fusion model levels is greatly improved compared with the single spectral model. In the data layer fusion rice origin identification model, the LS-SVM algorithm combined with SNV+2nd preprocessing method is the best model. The accuracy of the calibration set and validation set are 100% and 95.65% respectively, which is 2.17% higher than that of a single spectral optimal model. In the decision-level fusion identification model, the LS-SVM algorithm combined with the SNV+1st preprocessing method is the best model. The accuracy of the calibration set and validation set are 100% and 97.83% respectively, which is 4.35% higher than that of a single spectral optimal model. In the feature layerfusion origin identification model, the LS-SVM algorithm combined with SNV+2nd preprocessing method is the best identification model. The recognition accuracy of the calibration set and validation set are both 100%, which is 6.52% higher than that of the single spectral optimal model. The results show that it is feasible to use near-infrared spectroscopy, mid-infrared spectroscopy and Raman spectroscopy combined with chemometrics to identify rice origin, and the rice origin identification model based on Raman spectroscopy combined with LS-SVM algorithm is the best. The recognition accuracy of the three spectral information fusion model levels is greatly improved compared with the single spectral model. Among them, the feature level fusion method is more suitable for the data type of this fusion and can quickly and accurately identify the origin of Wuchang rice, Southern rice and Northeast rice. This study provides a new method for rapidly and accurately identifying rice-producing areas.