Research On Microexpression Recognition Technology In Intelligent Classroom Evaluation System

With the potential application value of facial expression recognition technology in disease diagnosis and teaching evaluation being explored,the technology has attracted extensive attention from many experts in different fields such as intelligent medical treatment and intelligent education.There are many good solutions for facial expression recognition technology in ideal scenarios.For real scenarios such as classroom teaching scene,facial expression recognition is still having severe challenges due to different problems such as small range of expression change,Presence of facial occlusion.To solve the problem of expression recognition in classroom environment,an improved residual network model based on shuffle attention mechanism is proposed.Based on this network model,a four-branch shuffle attention network model based on key region features is designed,and the recognition accuracy is better in the experiment of public data sets.The main work of this paper is as follows:(1)Aiming at the problems of expression recognition in classroom environment,such as small range of expression,unapparent feature of expression,small inter-class difference of some expressions,an improved residual neural network model based on shuffle attention mechanism is proposed.Firstly,shuffle attention module including spatial and channel attention mechanism is added to the residual network.The model can accurately extract significant facial features and improve the accuracy of expression recognition results.Then,the loss function was improved.Focal loss was used to reduce the impact of unbalanced sample distribution on performance of facial expression recognition model,and Center loss was used to solve the problem of some expressions features whose intra-class variation is large and inter-class difference is small.Finally,the proposed method is experimentally verified by using the public data sets CK+ and FER2013.The results show that the shuffle attention mechanism and the improvement of loss function can improve the performance of facial expression recognition.(2)To solve the problem of occlusion interference in facial expression recognition in classroom teaching scenes,A quadruplet shuffle attention neural network model is designed.Firstly,Dlib is used to locate the important areas of expression(eyebrow,mouth,nose),and separate these important expression regions from the original image,different network branches are used for feature extraction,and weights are assigned to the features of different branches according to the information entropy difference of the corresponding regions of each branch,and then feature information fusion is carried out.which reduces the interference brought by local occlusion to expression recognition,improving the robustness of the model.Finally,the method performance of robust is experimentally verified by using the public data sets FER2013 and RAF-DB,and the experimental results show that the proposed expression recognition method can still maintain a high accuracy of expression recognition under occlusion environment.(3)The expression recognition algorithm proposed in the paper has been preliminarily applied in classroom teaching evaluation.Firstly,the expression recognition model is trained on the self-built classroom expression data set,and the trained expression recognition model is used to recognize students’ expressions in class.The number of various expressions in different time periods is counted,and the emotional activity of students in different time periods is calculated according to the activity score of expressions.Students’ emotional activity is used to evaluate and analyze students’ learning state and classroom teaching effect.Finally,the evaluation results of classroom teaching are obtained.The experimental results show that the evaluation results obtained based on the expression recognition algorithm proposed in this paper are consistent with the actual classroom observation results.It shows that the classroom microexpression recognition algorithm proposed in this paper has potential application value in classroom teaching wisdom evaluation.