Emotion Analysis Of Students In Smart Classroom Based On Deep Learning

In the classroom,the emotional state of students is closely related to cognitive activities.The emotional state can reflect the state of learning,and it can also reflect the learning atmosphere in the classroom.Therefore,it is of great significance to accurately recognize the emotional state of students in the classroom.In the traditional classroom,the recognition of the emotional state of students is mainly through the teacher’s naked eye observation,which will affect the teacher’s attention in class.When there are many students,the teacher does not have the energy to observe the changes in the emotional state of each student.In online teaching,the interaction between teachers and students is limited,and the number of students may be larger.Teachers cannot observe the status of students in real time,which makes the teaching effect inferior to traditional classrooms.Therefore,how to use the machine to automatically recognize the emotional state of students is an urgent problem to be solved.This article applies related technologies in the field of deep learning to students’ emotion recognition,from the two aspects of students’ facial expressions and behaviors and gestures.By recognizing the facial expression and behavior posture of the student,the emotional state of the student is obtained.The research content of this article mainly includes the following aspects:Firstly,the development of deep learning related technology and related theories and research on student emotion recognition at home and abroad are combed,and a set of student facial expressions and behavior posture data sets are independently constructed.The facial expressions of the students are defined as: concentration,boredom,confusion,distraction and tired,and the behavior and posture of the students are defined as: look up,look down,look around,raise hand,play with phone,and stretch oneself.Secondly,an improved facial expression recognition method for students is proposed.The specific method is: on the basis of the Tiny＿YOLOv3 target detection algorithm,the attention mechanism is added to improve the convolution structure,and the GIo U loss is used to improve the loss function,and use the K-means algorithm to re-cluster on the self-constructed facial expression data set to obtain an anchor suitable for student expression recognition.Finally,after training,a model ER＿Tiny＿YOLOv3 suitable for student facial expression recognition is obtained.The experimental results show that ER＿Tiny＿YOLOv3 can accurately recognize the facial expressions of students,with m AP@0.5 reaching 0.79,Recall reaching 0.829,and F1 score reaching 0.671,and compared to Tiny＿YOLOv3,Precision has increased by 7.3%,and F1 score has increased by 4.2%.Thirdly,a behavior and posture recognition method based on the joint points of human bones is proposed.The specific method is as follows: First,use Gaussian filtering to eliminate noise,then use the target detection algorithm to detect the position of the student,then replace the Open Pose feature extraction network with Mobilenet series network,at the same time,use three 3x3 small convolutions with a residual structure to replace a 7x7 large convolution,and adopt a deep separable convolution form.and finally,the improved Open Pose is used to extract the coordinates of the bone joint points,and the bone joint point coordinates are classified by ST-SVM.Experimental results show that the student behavior recognition method based on human bone joint points can accurately identify the behavior and posture of students,with an accuracy rate of over 99%,and the frame rate is above 20.Finally,on the basis of the above research,the facial expression recognition results and behavior gesture recognition results are integrated at the decision-making level to make judgments on the emotional state of students.