Data-driven Optimization of Folk Dance Inheritance and Protection Strategies and Their Realization Paths

Abstract In this study, folk dance movement images are captured by visual image camera, the captured images are converted to gray scale images by averaging method, and Gaussian model and median filtering method are used to subtract the binarized image background and remove the noise in the images. After that, the neighbor frame feature processing module is proposed to draw on the attention mechanism for the temporal feature extraction of folk dance movements, and the temporal features of different scales are extracted under the multi-branch temporal feature processing module. The MI module and AF module are added into the folk dance action recognition network to form a new skeleton-based multi-channel topological spatio-temporal separated adaptive graph convolutional network (MTSGCN). Example analysis shows that the action recognition accuracy of the MTSGCN model on the validation set (92.21%) is significantly higher than that of the baseline model (83.46%), and its recognition accuracies of nine single and double folk dance actions are improved by an average of 16.8% and 18.04% over the baseline model. In addition, this paper can integrate and optimize the resources and materials of folk dance cultural characteristics to form a comprehensive and systematic dance inheritance development model.