| With the continuous transformation of educational resources in the Internet era,online education has gradually developed and matured.Different from the traditional education model,the diversification of resources in the online education environment allows learners to choose courses they are interested in,which not only enriches the way of learning,but also further promotes the development of individualized and fair education.However,while online education brings convenience to learning,there will also be problems that teachers cannot accurately grasp students’ learning status and learning emotions because online education is separated from teachers’ face-to-face guidance.A large number of studies have shown that facial expressions are the direct expression of emotions..Therefore,it is possible to obtain the emotional state and psychological state of students in the process of learning by studying the learning expressions of students in online education,and further analyze the changes of students’ learning emotions,so that teachers can adjust teaching strategies in time to improve the teaching effect.With the widespread application of deep learning in recent years,convolutional neural networks have been widely used in image recognition processing because of their better generalization ability and robustness compared with traditional methods.Facial expression recognition research is also a research hotspot in the current society,and expression recognition has received attention in the field of education due to the convenience of data acquisition and efficient use.Therefore,this thesis starts from improving the feature extraction ability of convolutional neural network,and explores the method to make the model have better expression ability,and finally improve the accuracy of expression recognition in order to show a better recognition effect.The research is applied to the facial expression recognition of students listening to class in the online education environment,which conforms to the deep integration and development of education and information technology in the Internet age.Therefore,the main work of this thesis is as follows.(1)Based on the depthwise separable convolution operation,the parameter amount of the model can be reduced to improve the feature extraction capability of the model.Therefore,by merging the depthwise separable convolution with the classic Le Net-5 model,the depthwise separable convolution layer is constructed by A fusion depth separable convolution classical network model is introduced,and the Dropout mechanism is introduced to improve the The Le Net-5 model has low recognition accuracy when dealing with image expression recognition research.In order to enable the model to better perform feature extraction on the experimental data set,this thesis conducts face detection,scale normalization and data enhancement on three experimental data sets,CK+ data set,Oulu-CASIA data set and OL-SFED learning expression data set.The final experimental results show that the recognition accuracy of the improved model has been greatly improved,and the model expression ability has been effectively improved.(2)A two-channel separable convolutional expression recognition model based on attention mechanism is designed,which replaces the standard convolution with depthwise separable convolution,and uses convolutions with sizes of 5×5 and 7×7,respectively.The product kernel connects the activation function,the BN layer,and the two branches connected by the SE module to form a two-channel separable convolution module.The SE module is a typical channel attention mechanism,which can assign different weights to different channel features,strengthen the Effective features suppress invalid features to improve the feature extraction ability of the model,and the recognition accuracy is also improved,and the model has a good recognition effect on three experimental data sets. |
PDF Full Text Request