A Study On The End-to-end Students' Emotion Recognition System

The traditional teaching mode generally adopts the "one-size-fits-all" large-scale talent cultivation method,which has an obvious shortcoming that is: in the teaching process,the lack of interaction between teachers and students.When students are not motivated by teachers for a long time,the so-called "affective deficit" problem occurs,that is,students gradually lose their enthusiasm and enthusiasm for learning.On the one hand,"lack of emotion" will make students tired of learning;on the other hand,it will make teachers unable to timely grasp students' emotions and then timely adjust the progress of the course according to the changes of students' emotions,which will greatly reduce the effect of teachers in class.How to solve the problem of "lack of emotion" has become the focus of scholars' research.Aiming at the problem of "emotion deficit",this paper proposes an end-to-end student emotion recognition system which is used to assist university classroom teaching.This system uses deep convolution network to realize the students' facial expression recognition function,and abandons the time-consuming and laborious manual extraction of facial features in the traditional method of facial expression recognition,and has the "end-to-end" facial expression recognition function.The system can detect the students' emotional changes in the course in real time by analyzing the classroom monitoring video,and then feedback the students' emotional changes to teachers in the form of visual charts,and put forward targeted teaching suggestions to teachers,which can improve the teaching quality and efficiency.This paper first analyzes the function and performance requirements of the end-toend student emotion recognition system,and gives the design of the overall framework of the system,and then designs the core modules and auxiliary modules in the system.Then the software and hardware development environment of the system is introduced,and the code realization of the core modules such as the user graphical interface,the image acquisition module,the expression recognition module and the training process of the expression recognition module as well as the result of the training accuracy and loss function were given.Finally,two sets of experiments are designed to evaluate the system from two aspects of functionality and robustness.