As a cultural subject, traditional music and art are facing the technical problem of incompatible conversion forms in the inheritance and development of modern society. In this paper, we take “music score recognition – music visualization – cross-cultural transmission” as the research idea, and focus on constructing the application framework of AI technology. The detailed features of musical notes are extracted using Harr wavelet transform, and the note contour layer is processed by 1D convolution operation, and a multi-scale feature fusion CRNN framework is established for sheet music recognition. After completing the transcription of musical notes, the VGGish model is used to process the music data, and the t-SNE algorithm is used to visualize the music data, forming a cross-cultural output scheme for traditional music art. In the practical application of this scheme, the effect of music performance was recognized by 80.00% and above of the subjects, and the visualization effect was positively evaluated by 67.27% of the subjects, which is an effective development path for the traditional music art form to fit into the contemporary society under the AI technology.
