Research On Students' Classroom Behavior Recognition Based On Deep Learning

With the rapid development of computer technology in the field of education,more and more researchers are focusing on the classroom.In the teaching process,it consists of the interaction between students and teachers.In order to improve the quality of teaching,it is also an indispensable link to study students' performance in class.In traditional classrooms,teachers mainly learn about students' performance in the classroom by observation,but this method cannot provide timely and effective feedback to teachers.So this article studies how to use deep learning methods to identify students' classroom behavior.This paper uses convolutional neural networks and transfer-generating adversarial networks to learn image features,and then classifies students' classroom behavior.The main work of the paper is as follows:(1)Making the data set.There are no publicly available data sets on student classroom behaviors on the Internet.In this paper,2024 images of five kinds of behaviors of 100 students are collected to construct a database of student's classroom behavior recognition.The five behaviors are reading the blackboard,reading a book,sleeping,turning around,and playing with a mobile phone.At the same time,data preprocessing is used to expand the data to obtain multiple data sets.(2)Classroom behavior recognition based on convolutional neural network.Based on the classic convolutional neural network model structure,this paper improves it and designs a network structure suitable for identifying student behavior.The final network parameters are selected by comparing the design parameters to determine the network structure.The final experimental result reached 80.7%.At the same time,this model is compared with the traditional CNN network,Lenet-5 network and Alexnet network.The experiment shows that the network model of this paper is superior to the other three network models in the accuracy rate and operation time of students' classroom behavior recognition.(3)Classroom behavior recognition based on migration MARTA-GANs.In this paper,we use the migration learning method to transfer the Marta Gans generation confrontation model to this study,and use the model to extract image eigenvalues,improve the classifier of the model.The experimental results show that the accuracy of the improved model is 6.7% higher than that of the original network,and the stability and robustness of the improved model are better.