The identification of Chinese herbal medicines is a key issue in the field of traditional Chinese medicine. Based on the characteristics of Chinese herbal medicines, the classification of types, producing areas, and quality can be realized. However, traditional identification methods of Chinese herbal medicines mainly rely on manual identification methods, which requires a lot for identification personnel with low efficiency. To solve this problem, we study the intelligent method of identification of Chinese herbal medicines by using data of infrared spectroscopy characteristic. To solve this problem, this paper studies the classification of spectral characteristic data of Chinese herbal medicines from unsupervised and supervised learning. Firstly, an improved K-means clustering algorithm based on Gaussian distribution model is established for unsupervised spectral classification of Chinese herbal medicines. This method “over-classifies” the sample data by K-means clustering algorithm, and further classifies the data by Gaussian mixture model, thus realizing unsupervised classification of Chinese herbal medicines. Secondly, aiming at the supervised classification and recognition of Chinese herbal medicines, an improved discriminant analysis classification method based on Gaussian distribution is established to identify different kinds and producing areas of Chinese herbal medicines. Finally, we test our method on two sets of data with and without tagged information, with Chinese herbal medicines in two data sets identified respectively. The experimental results fully verify the effectiveness of the method, especially in the supervised identification of Chinese herbal medicines. We have proved the effectiveness of our designed model through the comparison of various methods and extensive tests.