Artificial intelligence technology is integrated into guzheng teaching to promote the development of guzheng music and cultural heritage. This paper mainly establishes a fingering annotation method based on pitch difference fingering model and priori fingering knowledge to improve the learning effect of intelligent guzheng teaching. In the process of recognition of fingers, a two-way network is used that can take into account previous information. The Hidden Markov Generative Model enhancement method is added to the training set to enhance the data features. A priori knowledge of fingering is also introduced to correct incorrect transfer relations. The results of the empirical analysis show that the accuracy of the individual fingerings in the audio evaluation exceeds 90%, and the frequency values obtained are compatible with the standard tone (750 Hz). Compared to the control group, the experimental group’s increase in playing accuracy before and after practicing amounted to about 85.87%, and their performance level and mastery were improved. There was a significant increase in the satisfaction of the four users with the model before and after practicing (p<0.01).