Research And Practice Of Online Classroom State Monitoring System Based On Expression Analysis

With the advent of artificial intelligence and data age,the Internet plus education model has been blowout.In early 2020,novel coronavirus pneumonia outbreaks led to further progress in online teaching mode.However,due to the existence of space-time separation and single line teaching mode,online classroom lacks of humanized communication.When students' learning emotions and behaviors are back to the learning classroom,teachers can not timely correct and give reminders,resulting in poor learning effect.Therefore,how to obtain students' learning behavior from the online classroom,analyze and determine their classroom state,and form an effective classroom state evaluation mechanism is an important basis for realizing scientific online education and improving learning effect.In order to solve the above problems,this paper studies the online classroom state monitoring system based on expression analysis.(1)This paper summarizes the research background and significance of online classroom state monitoring based on expression analysis,as well as the traditional face detection and face recognition methods.This paper summarizes the current research progress of face detection and face recognition at home and abroad.The key technologies of the system and the theoretical basis of face detection,face recognition,fatigue detection and convolution neural network are introduced.(2)Aiming at the disadvantages of traditional face detection algorithm,such as large computation,poor real-time performance and unable to meet the actual needs,an improved MTCNN model is proposed.The model improves the image pyramid preprocessing method of MTCNN,combines multi-scale image pyramid with multi template image sampling method,and proposes a Gaussian pyramid preprocessing method based on multi template.In order to reduce the computation amount and improve the expression ability,more information needs to be extracted in the image.The model improves convolution operation in traditional MTCNN model and optimizes the network structure of the model.In order to reduce the computation and obtain the face information in the image efficiently,Siam factor is added after P-Net and R-Net of MTCNN respectively.The output results of the sub network are screened to reduce the number of useless candidate boxes transmitted.Experimental results show that the model is advanced and effective.(3)Focusing on the research topic of online classroom state monitoring and aiming at the defects of current face recognition methods,a face recognition method based on improved AlexNet network structure is proposed.The improved AlexNet network is used for face recognition,and the idea of global average pooling is used to reduce the dimension of the image,which effectively reduces the amount of parameters and over fitting phenomenon.Taking into account the cohesion within class,adjusting the variance within class and between classes,the Triple loss function is improved,and the original Softmax function is replaced by the improved triple loss function.Experiments show that the performance of the improved model is better and the efficiency of face recognition is improved.(4)Aiming at the problem of class state classification,this paper designs two classification models: CNN-SVM and VGGNet-O.CNN-SVM model extracts the features of eyes and mouth by using the method of cascade convolution neural network,and calculates the blink frequency.Then,the mouth shape change value,eye change value,blink frequency and PERCLOS value are used as input,and SVM method is used for classification.The VGGNet-O classification model extracts and classifies features based on the improved VGGNet network.In addition,Oct Conv convolution unit is introduced into VGGNet-O model to divide the features into high frequency and low frequency features to reduce spatial redundancy.Finally,the two classification methods are tested and compared by using the data set.The average recognition accuracy of the two methods for the three classroom States is 90%,and the average accuracy of vggnet-o model is 3% higher than cnn-svm model,which proves the correctness and effectiveness of the classification model.(5)The detailed design of online classroom state monitoring system based on expression analysis.Firstly,the requirement analysis,design idea and workflow of the system are described.Secondly,according to the requirement analysis,the UML use case diagram of the system is designed,and the database E-R diagram and database table are designed.Finally,the design and practice process of web and server in the system are described in detail.