Dance movement emotion recognition integrating CNN and long short term memory networks

Dance, as a visual art, carries the different emotions of dancers through its movements. In response to the problem of low accuracy in traditional emotion recognition methods for dance action emotion recognition, this study proposes a dance action emotion recognition method that integrates convolutional neural networks and long short-term memory networks. It extracts features of dance action images through convolutional neural networks, recognizes actions utilizing long short-term memory networks, and then constructs a dance action emotion recognition model. Finally, by incorporating attention mechanisms into the recognition model, the key information of dance movements is focused. The experiment showed that the recognition accuracy of the research model reached 0.9472, with a root mean square error of 0.5124, significantly better than other models. In the practical application analysis of the proposed method, the recognition accuracy of the four different emotions of happiness, fear, relaxation, and sadness in dance movements reached 98.21%, 99.24%, 97.32%, and 98.49%, and was more practical than the comparative models. The above outcomes indicate that the research method can enhance the efficiency and accuracy of dance action emotion recognition, and provide a reference for subsequent scholars to conduct research in emotion recognition.