Classroom Student Concentration Evaluation Method Based On Machine Vision

With the development of education in China,the number of college students is increasing every year.There are a large number of students in college classes,so teachers can’t monitor students’ learning state in class.In order to improve the learning efficiency of students in class,find out student’s concentration level in class,promote teachers to improve the teaching method and improve students’ interest in class.This paper tries to use machine vision to automatically identify the behavior of students in class and obtain the student’s concentration level in class.Due to the largest number of university students in the classroom in class,crowded seats,most of the student body characteristics are usually obscured,which makes it is difficult to balance the computing speed and recognition accuracy,to solve these problems,the article proposes classroom behavior recognition method,respectively,including the classroom students’ learning behavior identification method and class student behavior recognition method.Firstly,a learning posture detection method based on pose estimation was proposed,and the combination of pose estimation network and coordinate classification algorithm was used to identify the two learning behaviors of students: looking up at the blackboard and looking down to read in class.After my research on the existing pose estimation Network,I decided to improve the High Resolution Network(HRNet).The Squeeze and Excitation Network(SENet)structure is used to improve HRNet.Experiments on COCO show that the m AP of the improved HRNet increases by 0.6%.A behavior classification network based on support vector machine(SVM)was designed by using the data of key points identified by the pose estimation network.Experiments on the dataset collected in class showed that the recognition accuracy of the two learning behaviors reached 90.1%.The abnormal behavior detection method based on object detection proposed in this paper uses the object detection network to detect students’ sleeping behavior and playing mobile phone behavior.After studying the existing object detection network in this paper,this paper decided to use YOLOv4(You Only Look Once)object detection network,and proposed to use attention mechanism to improve the object detection network.the object detection network was improved by emending the Convolutional Block Attention Module(CBAM)into the backbone of YOLOv4.Experiments on the common Pascal VOC data set show that the improved YOLOV4 increases m AP by 1.2%.testing on the dataset collected in the classroom show that the accuracy of the improved YOLOV4 network can reaches 92.5%.Finally,combining the identification of classroom learning behavior and abnormal behavior,a class attention evaluation method is designed.the method divides the degree of attention according to the learning behavior and abnormal behavior,and allocates the weight.Then,the matrix algorithm of fuzzy comprehensive evaluation is used to evaluate the degree of attention in class,so as to accurately deal with the fuzzy problem of degree of attention.