Study On The Recognition Method Of Students' Classroom Behavior Based On Deep Learning

Classroom behavior recognition is one of the hotspots in the field of education.In the classroom,teachers judge students' learning status by observing students' behavior,but this method is time-consuming and labor-intensive,and cannot meet the needs of largescale classroom analysis in a smart learning environment.With the rapid development of deep learning technology in text processing,speech recognition,target detection,etc.,it is of great significance to explore the use of deep learning technology to identify students'classroom behavior.This paper focuses on students' classroom behavior in actual classrooms,and proposes to apply deep learning technology to classroom behavior recognition to monitor and analyze students' daily classroom behavior.This research has important significance for improving the quality of classroom teaching and improving the direction of teachers' teaching.Deep learning has great potential in the extraction of target features.This paper combines it with transfer learning technology to provide a research direction for the identification of student behavior in college classrooms.This paper works as follows:1?Build a database of student classroom behavior.For now,there is no public data on student classroom behavior.Therefore,through the observation and analysis of students!,classrooms,2083 images containing 9 types of behaviors were finally selected,including handing-up,falling,reading-books,listening,taking notes,fighting,playing-phones,talking-about and downing-table.2?Data preprocessing.Due to the complex background in the classroom and the small number and large number of individual students,the model obtained after training has the problems of low fault tolerance and low recognition rate.Therefore,Yolo-v3 is used to determine the position of the target object,and the image is cropped to reduce the interference caused by the background of the image.In addition,the dataset is augmented in various ways to increase the fault tolerance of the model.3?A dual network-based recognition model of students' classroom behavior is proposed.In this paper,the ResNet50 and VGG16 models trained on ImageNet are used,and the goal of recognizing students' classroom behavior is achieved through the dual-network model fusion framework.At the same time,the difference of network parameters will also affect the performance of model recognition.Experiments show that using the dual-network model structure,the average recognition accuracy can reach 92.3%,and the partial behavior recognition rate is as high as 98.2%,which is 2-5 percentage points higher than the recognition accuracy of the single-network architecture.Through a large number of experiments,it is proved that the stability and robustness of the improved model are better.4?A recognition model of students' classroom behavior based on ViT network is proposed.By adopting transfer learning technology,the trained ViT network model is transferred to the research of this paper,and the network structure is designed to adapt to the recognition of students' classroom behavior.In addition,appropriate parameters are selected through network training.Experiments show that the transformer model,which is mostly used in the field of natural language processing,can be used for individual detection and behavior recognition of students in complex scenarios by changing its network structure,and the accuracy rate reaches 92.3%,which is higher than the traditional network model.