The analysis of dance teaching system in deep residual network fusing gated recurrent unit based on artificial intelligence

The purpose of this study is to investigate how deep learning and other artificial intelligence (AI) technologies can be used to enhance the intelligent level of dance instruction. The study develops a dance action recognition and feedback model based on the Graph Attention Mechanism (GA) and Bidirectional Gated Recurrent Unit (3D-Resnet-BigRu). In this model, time series features are captured using BiGRU after 3D-ResNet is inserted to extract video features. Lastly, GA dynamically modifies the node weights to maximize action recognition performance. According to the experimental results, this model's F1 score is 85.34%, and its maximum accuracy on the NTU-RGBD60 datasets is more than 5% greater than that of the current 3D Convolutional Neural Network (3D-CNN) baseline algorithm. In addition, the model shows high efficiency and resource utilization in test time, training time and CPU occupancy. The research shows that this model has strong competitiveness in dealing with complex dance action recognition tasks, and provides efficient and personalized technical support for future dance teaching. Meanwhile, the model provides a powerful tool for dance educators to support their teaching activities and enhance students' learning experience.