| The innovative application of smart education driven by big data and artificial intelligence has become an inevitable trend to comprehensively promote the development of education digital transformation.How to effectively integrate the emerging technology with the education scene deeply and realize the precise and personalized teaching process has become one of the key issues to be solved urgently.As an important part of improving teaching quality,classroom teaching evaluation greatly promotes the transformation from traditional manual/semi-manual evaluation to automation,which is of great significance for realizing the intelligent development of teaching evaluation.At present,the vast majority of classroom behavior analysis and evaluation studies based on artificial intelligence technology focus on the process of students’ classroom learning,but the intelligent analysis and evaluation of teachers’ teaching behavior is still in the exploratory stage.For the research on intelligent analysis and evaluation of teachers’ teaching behavior,there are still problems such as lack of teacher behavior data set in real classroom scenes,difficulty in transferring existing behavior recognition algorithms and low accuracy of teacher’s classroom behavior recognition.This paper focuses on the research of teachers’ teaching behavior oriented to the application of smart classroom,explores effective methods of teachers’ teaching behavior identification based on deep learning,and provides technical support for carrying out personalized teaching assistance,intelligent evaluation and decision-making.The main research contents of this paper are as follows:(1)Construction of teacher behavior data set.This paper collected teaching videos of 100 teachers offline and online for 1000 class hours.Referring to S-T and TBAS teaching behavior analysis method,it classified 5 classroom teaching behaviors with obvious teaching characteristics,such as point to multimedia screen,point to blackboard,write on blackboard,operate multimedia,and gesture.The teacher’s classroom behavior data set containing 3032 video clips was constructed.(2)Teacher behavior recognition based on multi-stream graph convolution neural network.This paper proposes an improved multi-stream GCN teacher behavior recognition model based on the key bone features extracted from the HRNet human posture estimation model.By integrating the graph attention module mechanism and optimizing the connection relationship between nodes,the model can effectively integrate the information of joint,bone and movement information,and further enhance the performance of teachers’ behavior recognition.This study verifies the effectiveness of the proposed model through multiple comparative experiments on public data sets and self-built data set.(3)Teacher behavior recognition based on multi modal feature fusion.A multi modal feature fusion model is proposed to overcome the shortcomings of single modal bone features in the representation of teachers’ behavior features.In this model,three paths were designed to process RGB mode and bone mode data respectively,and the mode combining early and late fusion of features was realized through two-way transverse connection,which effectively fused image appearance features,key point spatial features and bone temporal motion features.The experimental results show that the method can effectively improve the recognition ability of confusing teaching behaviors,and the average recognition accuracy is 94.1%.(4)The design and development of teachers’ classroom behavior recognition system.Based on the proposed core algorithm,a teacher’s classroom behavior recognition system is designed and developed using Py Qt framework.The system supports core functions such as teaching video processing,human skeleton information extraction,teaching behavior recognition,and visual display of recognition results.It can realize real-time analysis of classroom teaching videos and provide effective means for intelligent teaching evaluation. |
PDF Full Text Request