Deep Learning-Based Analysis Of Teacher Classroom Teaching Behavior Research

The development of deep learning technology is currently advancing rapidly,and its applications are becoming increasingly widespread.As the technology continues to update and iterate,it has found its way into the field of education.However,in current research on intelligent education,the focus remains primarily on students,while supervision of teachers during the teaching process is often overlooked.This has resulted in the current evaluation of classroom teaching behavior being conducted manually by teaching supervisors,which is not only time-consuming and laborious but also subject to the influence of personal biases,making it difficult to provide an objective and long-term analysis of a teacher’s classroom teaching.However,by utilizing deep learning technology,it becomes possible to obtain real-time,fullprocess data on a teacher’s classroom teaching behavior.This data can be used to provide schools with a comprehensive understanding of a teacher’s long-term teaching dynamics,thus providing long-term,comprehensive,and objective data support for teaching evaluation work.The main work and research results of this article include the following aspects:(1)There is currently a dearth of datasets suitable for researching teacher classroom behavior detection,and challenges such as non-standard labeling and insufficient data volume persist.As a result,this thesis has developed a teacher classroom behavior dataset with ample data and standardized labeling to serve as the benchmark for our experiments.(2)After analyzing the current state of research on teacher behavior both domestically and abroad,and taking into account the real-time nature of teacher behavior detection in the classroom,this thesis conducted an analysis of existing object detection algorithms that perform well on teacher classroom behavior datasets.Ultimately,the YOLOX network model was selected as the foundation model.In addition,this thesis provides a detailed analysis of the network structure and basic operating principles of the YOLOX model.By conducting comparative experiments on teacher classroom behavior datasets,we confirm that the YOLOXs model outperforms the other three standard versions(YOLOX-x,YOLOX-m,YOLOX-l)in terms of speed and accuracy.Therefore,the YOLOX-s model was chosen as the base model,and subsequent optimization strategies were explored based on it.(3)Based on the analysis of YOLOX network model and the characteristics of teacher classroom behavior,Rep VGG network is used to replace the original CSPDarknet network in YOLOX model.In addition,analysis of the teacher classroom behavior dataset reveals the complex and varied environment,inconsistent detection target scales,and uneven light distribution,which makes it difficult for YOLOX model to focus on extracting key features from such image data.Therefore,ECA attention module is introduced based on YOLOX model.Moreover,in order to avoid losing too much semantic feature information during feature fusion,ASFF module is fused after the Neck part of YOLOX model.Finally,through comparative experiments,the results show that compared with the original YOLOX model,the m AP of the proposed model is improved by 7.91%,and the inference speed is increased by approximately24% on the teacher classroom behavior dataset.On the VOC2012 public dataset,the m AP is improved by 6.83%,and the inference speed is increased by approximately 28%.Finally,the proposed model is compared with several currently performing models through comparative experiments,and the results show that the overall performance of the proposed model is better than other detection models on both teacher classroom behavior dataset and VOC2012 public dataset.