Representation Of Attention And Learning Sentiment Based On Time-series Images In The Classroom

In the context of education reform,building smart classrooms is a means of breaking through the limitations of traditional education methods and actively creating new forms of education.In the vast amount of data available on the smart classroom platform,studying the correlations between the data can help to provide a continuous understanding and deeper study of learning emotions and to make accurate judgments about new learning processes.In this paper,we analyze the learning time sequence image data from the smart classroom platform and construct a method for characterizing students’ attention and learning emotions in the classroom.1.To address the problem that the existing algorithms for characterizing students’ learning emotions in classroom scenarios do not take into account the different learning emotions in different attention states of the same expression resulting in low accuracy,a method based on Dempster-Shafer evidence theory fusion analysis of classroom attention and learning emotion representation is proposed,and an expression dataset applicable to real classrooms is constructed on this basis.The method first combines a channel-based cross-fusion Transformer and multi-layer sampling feature extraction to construct a ResNet-based expression recognition model.Finally,the Dempster-Shafer evidence theory was used to analyze classroom attention using head posture data and then fused with the facial expression temporal data to construct a model for the representation of learning sentiment during the lecture time of knowledge points.The experiments show that the proposed learning sentiment analysis method can effectively characterize learning sentiment,with the expression recognition model achieving an accuracy of 73.58% on the FER2013 dataset.2.To address the low accuracy of existing methods of classroom attention state analysis,a Bayesian probabilistic fusion-based method for analyzing classroom attention and learning emotion representation is proposed.The method is based on the Fine-Grained Structure Aggregation network framework,combining multiplex feature extraction and Squeeze-and-Excitation attention mechanism to optimize the head posture estimation model.Secondly,the optimal estimation model is used to obtain head Euler angular time series data and to analyze students’ classroom attention using the transformed relationship between head Euler angles and classroom attention positions in conjunction with Bayesian probability theory.Finally,the classroom attention state and facial expressions are integrated to portray learning sentiment.Experiments show that the proposed attentional analysis algorithm has better performance than existing analysis methods,where the MAE of the head pose model reaches 5.00.The proposed method is validated by using videos of students’ lessons and assessment results in a classroom in a smart classroom.The experiments show that the proposed method can analyze students’ attention and learning emotions more effectively than the current research,provide data support for teachers’ process evaluation of students,and help teachers to adjust their teaching methods and promote personalized teaching.