Abstract: By recognizing the facial expressions, the emotional state of learner can be judged and their learning effect can be analyzed. Due to the persistence and timing of facial expressions, the sequence of facial images is adopted as the object of facial expression recognition. The Long Short Term Memory Network (LSTM) and VGGNet are combined into a VGGNET-LSTM model. On this basis, facial expression recognition is carried out, which significantly improves the accuracy of recognition. Based on the transfer learning method, VGGNet is transferred to the learning expression data set after being pre-trained through the basic expression data set CK+ and avoids the defect of insufficient data in the learning expression data set and solves the problem of overfitting the model.

Key words: expression recognition, academic emotion, image sequence, transfer learning