Lily bulbs, the underground metamorphic stems composed of thick scales grown by perennial herbaceous bulbous plants of the lily family Liliaceae, a typical medicinal and edible homologous crop. It is rich in nutrients and has anti-tumor, antidepressant, hypoglycemia, and improves immune functions. The market prices of lily bulbs from different origins are quite different. The traditional evaluation methods that rely on artificial experience and sensory are highly subjective and have poor certainty, making it difficult to be widely used in modern production links. Advanced detection methods based on chemical inspection methods are time-consuming and expensive and it are difficult to meet the requirements for origin identification. Raman spectroscopy is a vibration spectrum based on inelastic scattering, which can achieve fast and accurate non-destructive testing. Combining Raman spectroscopy with machine learning algorithms, a classification model of the three most widely distributed lily bulbs in China (Lanzhou lily, Yixing lily and Longya lily) was established. Observing the characteristic peaks of 479, 870, 942 and 1 606 cm-1 on the matrix spectrum, a non-destructive testing method based on the component content of Raman spectroscopy to determine the place of origin and evaluate the quality of lily bulbs is proposed. First, the traditional method is used to collect the spectrum of the lily bulb sample. After the spectral data is preprocessed, the artificial prior method is used to extract the representative substance of the lily bulb and determine the characteristic peaks. Then the principal component analysis and the t-distribution random neighborhood embedding method are used to reduce the dimensionality. Extract spectral data features. The data features obtained above are applied to support vector machines, decision trees and random forest algorithms. The experimental results show that these classification models all show ideal classification accuracy on the same test set. Among them, the model’s accuracy based on principal component analysis and decision tree algorithm reached 91.7%。 The model’s accuracy based on t-distribution random neighborhood embedding and support vector machine is 93.7%, and the accuracy rate of the model combining the principal component analysis and random forest algorithm is as high as 95.8%. In summary, this method can provide on-site rapid identification and identification of the origin of lily bulbs, improve the accuracy of the quality assessment in the modern production process, and provide a reference for the identification of the origin of modern production and the quality analysis of lily bulbs.