To prevent the adulteration of pure milk and optimize the detection methods of pure milk, a recognition scheme for pure milk based on near-infrared spectroscopy is proposed in this paper. Firstly, a Fourier near-infrared spectrometer was adopted to obtain the near-infrared spectroscopy signals of different pure milk products from the same brand within the wavelength range of 4 000～10 000 cm^-1. Since the obtained near-infrared spectroscopy signal data is relatively redundant, this paper utilized a principal component analysis algorithm to extract the feature information of the near-infrared spectroscopy signal data within this range to improve the recognition efficiency of pure milk. Four principal components with larger contributions were extracted to obtain the training and testing sample data. Then, the back propagationneural algorithm was employed for preliminary training and testing of the obtained sample data. The test results show that the BP neural network algorithm combined with principal component analysis improves recognition efficiency of pure milk and achieves an accuracy of up to 95%. To further improve the algorithm's accuracy, the particle swarm optimization algorithm was added to the proposed pure milk recognition scheme to optimize the weights and thresholds in the BP neural network. Additionally, a new dynamic decreasing inertia weight factor function was proposed in the particle swarm optimization algorithm for the inertia weight factor. Experiments show that the accuracy of the proposed intelligent recognition scheme for pure milk can be increased to 100%. Therefore, the intelligent recognition scheme for pure milk based on near-infrared spectroscopy can accurately and effectively identify the types of pure milk.