Research on Emotional Design of Hanfu Appearance Based on Machine Learning and Style Transfer

This study investigates the emotional design of Hanfu using BP networks. Different styles of Hanfu are selected as research samples, and their styles are analyzed and deconstructed to determine the weight of structural elements in emotional evaluation. Secondly, emotional vocabulary is collected from users, and representative emotional words are selected through clustering. The Riket scale is used to determine the emotional evaluation value of the research samples. Finally, the method of rapid style transfer is used to extract sample style features, which are concatenated with structural features to obtain sample feature vectors. A BP neural network is used to establish the model relationship between sample features and emotional words. The network is trained using PyTorch and validated using a validation set.