Keywords: 
• micro-expression recognition /
• spatial-temporal features /
• LRCN /
• LSTM /
• Education evaluation

Abstract: With the rapid development of machine learning and deep neural network and the popularization of intelligent devices, face recognition technology has developed rapidly. At present, the accuracy of face recognition has exceeded that of human eyes. At the same time, the software and hardware conditions of large-scale popularization have been available, and the application fields are widely distributed. As an important part of face recognition technology, facial expression recognition has been widely concerned in the fields of artificial intelligence, security, automation, medical treatment and driving in recent years. Expression recognition, as a hot topic of human-computer interaction, involves Informatics, psychology and so on, and has a good research prospect in teaching evaluation. Among them, micro-expression is a kind of short-lived facial expression that human beings unconsciously make when trying to hide some emotion, which has great research significance. Different from the general static facial expression recognition, micro expression recognition not only needs to extract the spatial feature information of facial expression deformation in the image, but also needs to consider the time motion information of the continuous image sequence. In this paper, for the problem of static expression features lacking time information, which cannot fully reflect the subtle changes in expression, etc., facial dynamic expression sequences are used to fuse spatial features and temporal features, and use neural networks to provide good features in the field of image classification. Expression sequences are processed, and a micro-expression recognition method based on S-LRCN is proposed. Firstly, the micro expression data set is selected to extract the facial image sequence, and the transfer learning method is introduced to extract the spatial features of the expression frame through the convolution neural network model of pre training, so as to reduce the risk of over fitting in the network training, and the extracted features of the video sequence are input into LSTM to process the time-domain features. Finally, a small database of learners' expression sequences is established, and the method is used to assist teaching evaluation.