Raman spectral analysis technology is used in the field of drug analysis. The traditional Raman spectral analysis method relies on the artificial expertise and testing conditions, which consumes a lot of time and manpower. To solve this problem, a method combining Raman spectroscopy and convolutional neural network (CNN) in deep learning is proposed in this paper, which can be more efficient and convenient for drug analysis. The Raman spectra of 4 kinds of drugs were collected in the experiment. The asymmetric least square method (ALS) was used for baseline correction and S-G smoothing filtering, and the original Raman spectra were preprocessed by Max-Min standardization. Six data enhancement methods, including the method of mixing multiple Dirichlet spectra, were combined to generate new Raman spectra to expand the data. The data enhancement method improved the fitting ability and expression ability of the model. Principal component analysis (PCA) method was used to reduce the dimension of the original Raman spectrum measured, and the spectral data was used as the input to build a one-dimensional convolutional neural network model to predict drug categories and realize drug classification, providing an intelligent method for drug analysis.