Student Classroom Emotional Analysis Based On Face Recognition

In the current domestic education and teaching system,students are the main body of the classroom,their emotional state of learning in the classroom is an important indicator to measure the learning result,and it is also an important basis for evaluating the quality of teaching.Currently teacher mainly obtains the real-time listening status of students through classroom observation and questioning.Using computers to recognize the facial expressions of students automatically obtains the real-time emotional state of students when they are in class,provided new idea for the current smart teaching and is a new method to realize smart teaching.This essay mainly studies how to obtain the emotional state of students when listening in class through facial expression recognition technology.To complete this work,it is necessary to use Face Detection Technology to locate the position of the face in the image,and then pass the detected facial image to the subsequent classifier for further recognition.The main research contents of this paper are as follows.Firstly,in order to improve the accuracy and efficiency of the Face Detection,this article has done the following three aspects of work for the MTCNN model.First,a dynamic minsize algorithm is designed to dynamically control the minimum face size of the model according to the size of the input image,which reduces the number of iterations of the image pyramid and improves the calculation speed of the model;Second,the use of deep separable convolution to replace the standard convolution structure in the P-Net and R-Net models,effectively reduce the amount of model parameters and calculations.Third,an O-Net model with a densely connected structure is designed.Through the reuse of different layer features of the image,the recall and accuracy of MTCNN model in the verification set are improved by 2.39%and 1.65%respectively.Improved the performance of MTCNN from the two directions of speed and accuracy.Secondly,the class emotion types that need to be identified are designed,and collected corresponding facial expression data samples.After consulting related literature on emotion recognition,the emotional state is divided into positive and negative states through questionnaires and offline interviews with teachers and students in school.The positive state includes listening,appreciation,and doubt,and the negative state includes resistance and inattention.And organized some personnel to collect tens of thousands of facial expression data samples containing five emotion categories for subsequent training and verification of emotion recognition models.Then,an expression classification model based on attention is designed.Human faces often contain a lot of information.In daily life,people generally can better distinguish specific expression types by observing some areas of the face.That is to say,different facial regions contain different emotional information.According to this characteristic,this paper introduces the attention mechanism into the classroom emotion recognition,and realizes a spatial attention model that effectively extracts the features of the key areas of the face.Experiments on the collected data sets showed that the four network structures of AlexNet,VGG16,GoogLeNet,and ResNet18,after adding the attention model,improved the recognition accuracy by 0.35%to 1.95%,which verified the effectiveness of the model.Finally,on the basis of completing the above work content,this article uses Python language to implement a easily manipulated classroom emotion recognition system.