Detection And Analysis Of Students’ Behavior In Classroom Based On Deep Learning

In recent years,intelligent education has gradually moved from theory to campus,and intelligent identification of students’classroom behavior has become increasingly important.In the course of education,students’behavior on the class is an important reference to evaluate student’s learning efficiency,to adjust teacher’s educational strategy,and to improve teaching management efficiency.Therefore,using intelligent methods to automatically identify students’classroom behaviors has gradually become the focus of educational research.This article focuses on students’classroom behavior in different teaching scenarios,and proposes to apply deep learning to the identification of students’classroom behavior to achieve automatic perception of classroom teaching effects,providing evidence for students to self-evaluate,teachers to adjust teaching methods,and improve management efficiency.This article takes teaching scene objects or students with certain behaviors as the research object,and uses object detection in computer vision as a method to study students’classroom behavior.The specific research content is as follows:(1)Establish a teaching scenario goal and student classroom behavior dataset—POBC~2(Pedagogical Objects and Behavior in College Classroom)dataset.Aiming at the problem that there is currently no publicly available dataset for student classroom behavior,this article proposes a dataset suitable for teaching objectives and student classroom behavior research by referring to classic behavior datasets and other research on student classroom behavior.This dataset is derived from real university classroom scenarios,covering 10 common types of student classroom behaviors and goals,and contains a total of 137960 sample labels.(2)Based on the Bi-Directional Feature Pyramid Network,a Yolov5-Bi FPN network model is constructed to detect and identify students’classroom behavior.In this paper,YOLOv5(You Only Look Once)is selected as the primary object detection benchmark model for this study.In order to solve the problems of insufficient feature fusion and poor performance of fine grained features under complex conditions,this paper proposes a new detection model—YOLOV5-BIFPN.This model can effectively improve the feature representation and fusion capabilities of YOLOv5.The results show that YOLOv5-Bi FPN has better detection performance,with a detection accuracy of 82.7%and an improvement of2.4%.(3)Aiming at the problem of difficult detection of small goals,a YOLOv5-Bi FPN-Sdet model is constructed to detect and identify students’classroom behavior.The dataset constructed in this article contains a large number of small objects,and the detection effect of YOLOv5 and YOLOv5-Bi FPN on this dataset is not ideal.By adding a detection layer for small objects to the network layer that outputs the large-scale feature map,the goal of improving the model’s ability to locate small objects is achieved.Finally,compared with the experimental results of other models,the results show that YOLOv5-Bi FPN-Sdet has the highest detection accuracy of 88.6%,which is 8.3%higher than the original model.Then,this model is applied to real classroom data to calculate the frequency of occurrence of various behaviors of class students,and analyze the class students’listening status and performance in the classroom.