Facial Expression Recognition Of Teenagers Based On Transfer Learning

The online education industry has received more and more attention due to the epidemic.Although online education has solved the space problem of teaching to a certain extent,the lack of real-time interaction between students and teachers affects the efficiency of teaching and also greatly affects the teaching efficiency.Is related to the professional level of teachers and even online education institutions.To this end,this thesis proposes a solution to this problem,which analyzes the current learning state of students by identifying changes in their facial expressions.At present,expression recognition generally uses convolutional neural networks,and its extremely high recognition ability has also been widely recognized.For convolutional neural networks,higher recognition capabilities depend not only on better models,but also on the data set.The requirements are also higher,and higher-quality data sets correspond to higher collection and production costs.For researchers,the samples of the current public data sets are all generalized,but there are not many facial expression recognitions for teenagers in online education scenarios.In response to the above problems,this thesis combines transfer learning with convolutional neural networks,and uses self-made data sets for transfer training,and finally forms a recognition model with strong feature learning ability for teenagers’ basic expressions.In this thesis,by understanding transfer learning and convolutional neural networks,collecting relevant theoretical knowledge and development status,combining the research direction of the subject to find and design a suitable network model and implement it.This subject mainly has the following aspects of work content and results.(1)Analyze the demands of teachers under online education and clarify the types and characteristics of facial expressions required for this topic.The current mainstream data sets do not distinguish ages and there are various types of facial expressions(6-15 types).Online education is aimed at student groups and There are few types of expressions(3-5 types).Although the data samples of the student population are a subset of the mainstream data set,because the data distribution is random and difficult to distinguish,this thesis will collect relevant materials on the Internet and make a data set for these problems.To satisfy the use.(2)Aiming at the research problem and the characteristics of the data set,this thesis proposes a solution to transfer learning,using the powerful generalization ability of the convolutional neural network and the public data set to complete the training of the generalization model,and freeze the migration training for each layer separately In the experiment,find the network layers where the model outputs general features,then freeze these network layers and use the personal data set for migration training.The results are verified,which can effectively identify the data categories in the personal data set and significantly reduce the training time of the model.(3)The joint distribution adaptive method is added to the network model,and compared with other classic models,it solves the problem of poor migration effect caused by the difference in sample types between public data sets and personal self-made data sets,and improves The robustness and flexibility of the model are improved.This thesis elaborates the above research content in detail,and the experimental verification on the public data set and the self-made data set shows that the experimental model proposed in this thesis can effectively recognize the facial expressions of teenagers.