In order to improve the analyzability of intangible cultural heritage dance in inheritance, creation and digital communication, this paper takes Yushuyi dance as the object and proposes an intelligent analysis method for dance innovation mechanism identification. This paper constructs a multi-source data system that integrates action skeleton, long-sleeve trajectory, rhythmic audio and cultural semantic tags, and designs a multi-modal fusion model to jointly model symbolic, prosodic and improvisational. The experiment uses 8:1:1 data division to complete the training and testing in a unified environment. The results show that the action recognition accuracy of the proposed model reaches 91.52%, Macro-F1 is 90.36%, and Top-3 coverage is 96.08%. In the impromptu-enhanced sample discrimination, the AUC reaches 0.918. The research shows that computer technology can provide quantifiable and interpretable technical support for the style identification, creation transformation evaluation and live inheritance of intangible cultural heritage dance.
