METHOD AND DEVICE FOR TRANSFER LEARNING-BASED FACIAL EXPRESSION RECOGNITION USING WEIGHTED CLUSTER LOSS

摘要

Disclosed are a face expression recognition method and a face expression recognition apparatus based on transfer learning that learn using a weighted cluster loss. The transfer learning-based facial expression recognition method according to an embodiment of the present invention uses a weighted cluster loss function to prevent the SE-ResNet-50 model from being biased and learned when an unbalanced dataset is used. calculating; updating the SE-ResNet-50 model by using the weighted cluster loss function; training the SE-ResNet-50 model as a final model by transfer learning using the weighted cluster loss function and the AfftecNet data set; and injecting facial data obtained from the subject into the final model to recognize the subject's facial expression.