A Study On Measuring Learning Behavior Engagement In Elementary School Classrooms Based On Object Detection Technology

Learning engagement is a key factor affecting learners’ learning outcomes,and learning behavior engagement is an important component of learning engagement.Process measures of students’ learning behavioral engagement in authentic classroom settings are important for reflection and improvement of teaching and learning.There are significant differences in the cognitive development and adaptive development levels of elementary school students at different grade levels,and the classroom behavioral performance of elementary school students is richer and has high measurement value.The emergence of target detection technology can significantly improve the efficiency of capturing and identifying classroom learning behaviors.The following problems exist in related studies: the classification of classroom learning behaviors is relatively simple and lacks theoretical basis;there is a lack of publicly available and valid classroom student behavior data sets;the recognition effects of different algorithmic models vary greatly for complex real classroom environments;the studies mostly focus on the improvement of algorithms without corresponding tools for automated measurement and analysis systems;and there is also a lack of application analysis of learning behavior data from actual classrooms.To address the above problems,the main work of this study is reflected in the following five aspects:(1)Building a framework for measuring classroom learning behavior engagement.Based on the theories of learning engagement,learning participation model,Bloom’s cognitive model,and ICAP,this study designed a "three-level" model for measuring classroom learning behavior engagement in three dimensions:ineffective learning,relatively shallow learning,and relatively deep learning.(2)Constructing an elementary school classroom learning behavior data set.Real daily classroom videos of an elementary school from Grade 1 to Grade 6 were obtained.The process of constructing the dataset mainly includes five steps: data collection,data cleaning,image extraction,data enhancement,image annotation and data division.(3)Optimizing the effectiveness of algorithmic recognition of learning behaviors in elementary school classrooms.The changes of different environments such as classroom lighting are simulated by linear transformation and histogram regularization to increase the amount of data while improving the generalization ability of the dataset to different scenes.Comparing YOLOv3,YOLOv5,and Faster R-CNN target detection models.the YOLOv3 algorithm model works best,and the average accuracy rate can reach97.47%.(4)Develop an automated identification and data analysis system for classroom learning behavior engagement.Using Python,Py Qt5,Sqlite3,Html and other technologies,we designed and developed a classroom learning behavior engagement analysis system,which can automatically monitor and record classroom learning behavior data.(5)Analyze the engagement of classroom learning behaviors in elementary schools from multiple same dimensions,such as whole school,lower high school grades,different courses in different classes,and learning alerts,and provide suggestions for improvement of teachers’ teaching and students’ learning.