Recognition Of Student Learning Expressions In Online Classroom Based On Deep Learning

Facial expression recognition is an important part of emotional computing and is widely used in education,medicine,human-computer interaction,and other fields.Recognizing students’ facial expressions can be used as an auxiliary teaching method to promote the quality of teaching and make it more effective.With the rapid development of computer vision,facial expression recognition is a key research topic that is gradually being extended to various fields.Due to the separation of time and space between faculty and students,emotional communication between teachers and students is difficult to carry out effectively,which affects teachers’ judgments of students’ academic emotionality and the ultimate effect of instruction.Therefore,recognizing students’ faces in online classes can help analyze their academic moods and promote the effect of classroom instruction.Convolution neural networks have attracted much attention in recent years.Therefore,it is necessary to study the application of convolution neural network in facial expression recognition system.Convolution neural networks can obtain automatic features by convolution checking images,which are better and more robust than traditional machine methods.On this basis,the face is detected and located with depth network so that the face recognition task based on depth model can be accomplished.This thesis has several main aspects:(1)The VGG16 network model is improved to solve the problem of a large number of parameters and slow operation.By changing the size of convolution kernel,increasing the number of convolution nuclei and increasing the normalization layer,we can improve the feature extraction ability of the network.Finally,we use the method of increasing batch normalization layer to improve the feature extraction ability of the network.To verify the validity of the improved VGG16 network model,simulation experiments were performed on five datasets —Fer2013,CK +,JAFFE,Oulu-CASIA and OL-SFED —and data enhancement was performed by flipping,rotating,and translational images.(2)In order to further improve the performance and feature expression of VGGG16 network,we propose a method to increase residual attention module to improve the role of network model in feature diagram processing,accelerate network convergence,and increase accuracy and speed of network model.After testing attentional VGG16 network model in four facial expression datasets —Fer2013,CK +,OL-SFED and JAFFE —we achieved better facial recognition than original VGG16 network model.Then after experimenting with some same networks we confirmed that attentional VGG16 network was more accurate in facial expression recognition than original VGG16 network in four facial expression datasets.