Student Classroom Behavior Recognition Based On Deep Learning

With the vigorous development of intelligent campus construction,the construction of informatization and networking in colleges and universities has moved from digitalization to intellectualization.In the process of teaching,students' classroom behavior has important reference significance for the development of teaching activities and the adjustment of teaching strategies.In traditional classroom teaching,students'behavior recognition is mainly realized by manual observation,but this method is not only cumbersome but also time-consuming.Obviously,it can't meet the needs of the big data in education.Therefore,it is an urgent problem to explore how to recognize students' behavior automatically by machine.This paper mainly adopts the way of transfer learning and makes use of the powerful feature learning ability of deep learning model to explore effective ways to solve the problem of automatic identification of students'classroom behavior.The main work of this paper is as follows:(1)Building data sets.At present,there is no public image database of students'behavior.In this paper,2100 images of seven kinds of behaviors of 300 students are collected to construct a database of students' classroom behavior recognition.Specific classroom behaviors include reading,sleeping,raising hands,writing,listening,standing and looking around.(2)Data preprocessing.In this study,in order to improve the effect of model training,the training set was preprocessed,including student detection and data enhancement.Student detection uses Yolo_v3 to detect the location of students in the image,and the image is clipped.In the process of data enhancement,this paper uses 12 ways to expand the training set.(3)Classroom behavior recognition based on deep learning.In this paper,ResNet50 network trained on ImageNet is used as the pretraining model,and the deep model is trained by the way of transfer learning for students' behavior recognition.At the same time,the effect of main parameters on recognition performance is also discussed through a large number of experiments.Finally,this paper develops a demonstration system of students' classroom behavior recognition based on deep learning for demonstration model training and testing.