In order to achieve accurate classification of watercolor pen ink, 60 samples of watercolor pen ink from 15 series of 3 brands are tested by infrared spectroscopy in this work. First, after preprocessing such as smoothing and correction, root mean squared error is used to determine the optimal wavelet transform compression times, and the purpose of reducing the complexity of the operation is achieved after compression. Then, H?lder exponent is used to extract 30 characteristic waves of 3 brand samples, which are imported into the input layer of artificial neural network as input variables. The training set, validation set, and test set are assigned to train the model, and the final classification accuracy of the model is 83.3%. Finally, receiver operating characteristic (ROC) curve is drawn, and it is found that the classification accuracy of the second kind of samples is higher than that of the other two types of samples, which realize the pattern recognition of watercolor pen ink types.