Research On Automatic Recognition Of Student Engagement Based On Computer Vision

At present,the country is paying special attention to artificial intelligence,and has issued relevant policies to help artificial intelligence + education.President Xi Jinping emphasized the need to actively promote the deep integration of artificial intelligence and education,promote educational reform and innovation,give full play to the advantages of artificial intelligence,and accelerate the development of education that accompanies everyone’s life,Equality-oriented education for everyone,education suitable for everyone,more open and flexible education.The student engagement problem has always been a very common and important issue in education and teaching,so it is necessary to identify the student’s learning engagement,which not only helps teachers to understand the student’s participation in order to intervene in time,but also helps students to reflect on their own learning and promote their in-depth participation in the learning process.By consulting the relevant literature at home and abroad,the author clarifies the current situation of artificial intelligence(computer vision)in the recognition of student learning engagement,and also learns about the convolutional neural network and its typical image recognition classification neural network(Vgg16,Vgg19,Inception v3,Res Net50 and Xception).Then,using the online student engagement open data set in Aditya Kamath conference paper to carry out model training and learning,the recognition classification performance of these models on student engagement is compared by recording data such as the participation accuracy and training time overhead of five typical image classifications,so as to filter out the optimal neural network mode.Then collect face images from the classroom video of the real classroom,and clean and classify the facial data,which is mainly divided into 3 levels(no participation,general participation,and active participation)to form a real classroom learning participation data set.Next,gradually add real classroom learning participation data in the training set of the online learning open data set,and use the previous experimental conclusions to compare and train the optimal image recognition classification neural network model Xception to train and learn,by recording the results of each Xception model training learning to verify the corresponding scenario data(real classroom studentengagement data)on the model’s recognition accuracy under the corresponding scene data,Finally,the influence of neural network model,participation level and corresponding scene data on student engagement recognition accuracy are discussed and summarized,and feasible and valuable suggestions are provided for the implementation of student engagement recognition,so as to provide reference for relevant research and practice.Based on the above research,this paper draws the following conclusions:(1)In the case of the same data set,different neural network models have an important influence on the accuracy of student learning engagement recognition.In this study,the Xception model is the best-performing engagement recognition model,followed by the Inception v3 model,followed by the Res Net50 model.In the enhanced data sets of levels 1and 3,the Xception model had a maximum accuracy of 98.48% of the recognition of online student engagement,with a training time of 37.6 minutes.(2)In the case of the same neural network model,the number of participation leveling has an important influence on the accuracy of students’ learning engagement recognition.In this study,the participation recognition accuracy of data sets with two participation rating is greater than or even greater than that of the participation rating classification of three.(3)In the case of the same neural network model,when the corresponding scene data is added to the basic training set,the model training results have an important influence on the accuracy of the identification of the corresponding scene.In this study,when more and more face data of real classroom students’ learning engagement put in the training set,the recognition accuracy rate of the Xception model for real classroom students’ learning engagement is increasing in the general trend.(4)In the case of the same neural network model,mixing data from a particular scene and other scenarios as training sets and single data under a particular scene as training sets,the model training results have an important impact on the accuracy of student engagement recognition in a particular scenario.In this study,whether it is a single-person data set in real classroom data,or its average accuracy,or a test set in an open data set,the student engagement recognition accuracy of mixed training sets is mostly higher than the recognition accuracy of a single training set.