Emotion recognition based on ResNet and transfer learning

Abstract

In the fields of psychology and artificial intelligence, emotion recognition is a crucial area of research. As an expression of body language, emotion recognition has a profound impact on our life and interpersonal relationships while existing models have the phenomenon of exploding or disappearing gradient indices due to the lack of information. To alleviate the aforementioned limitations, this paper further introduces the idea of residual jumping and transfer learning to conduct further research and exploration for the emotion recognition. Specifically, based on the RAF-DB dataset, the self-created model Awei and the pre-trained VGG model are combined. Numerous tests show that the suggested approach is effective. By reducing the amount of information lost during the convolution process, the residual idea can help with the issue of an exploding or vanishing gradient index brought on by an excessively long backpropagation. This will increase the correctness of the model. The pre-trained VGG model's addition improves the Awei model's parameter modification's efficacy and accuracy.