With the development of artificial intelligence technology, there are new ways to solve the problems of difficult personalized instruction, subjective movement evaluation and low teaching efficiency in dance teaching. This study explores the application of artificial intelligence in dance teaching. Based on constitutional neural network and long short-term memory network for dance movement recognition, diverse dance video data sets are constructed and data enhancement and transfer learning are carried out to improve the recognition accuracy and generalization ability of the model. The results show that the accuracy of the system in multiple dance movement recognition is more than 90%, which improves the scientific and objective teaching. Artificial intelligence technology has application potential in dance teaching, which provides strong support for the innovation of education mode.