Dynamic facial expression recognition model based on BiLSTM-Attention

It is common in most service industries to obtain customer satisfaction ratings in the form of questionnaires to improve services, an approach that suffers from sample acquisition difficulties, irrational distribution and subjective error due to customer willingness to participate. To address this issue, this article uses emotion recognition technology to analyze customer emotion expressions in videos instead of traditional questionnaires,but the task of this kind of dynamic facial expression recognition based on image sequences in the process of expression changes still faces huge challenges: traditional recognition methods ignore the relationship between expressions changes and keyframe information.Therefore, this paper proposes a network model based on BiLSTM-Attention structure in combination with VGG16. BiLSTM is used to obtain the contextual characteristics of pictures and capture the relationship between long-term emotional fluctuations. The application of attention mechanism assigns the importance of pictures representing short-term facial expressions of people, and make the model automatically learn information that is highly relevant to decision-making. Experiments show that this method can accurately determine people's comprehensive emotions over a period of time and has a good learning performance.