Learning Expression Recognition Based On Convolutional Neural Network

With the deep integration of Internet and education,online education is booming in China.The emergence of online education breaks the time and space limitations of traditional teaching,and to a certain extent,promotes the fair development of educational resources.Although online education has many advantages,its lack of emotion cannot be ignored.In the online education scene,due to the separation of time and space between teachers and students,teachers cannot effectively communicate with students emotionally,which affects teachers’ judgments on students’ academic emotions and the final teaching effect.Facial expressions are intuitive manifestations of person’s inner emotions.By observing facial expressions,one can obtain information about changes in person’s inner emotions.Therefore,the recognition and analysis of students’ learning expressions in online education can help to analyze the changes of students’ academic emotions and improve the final teaching effect.In recent years,convolutional neural network has been widely concerned.Convolutional neural network can automatically obtain image features through convolution kernel,which has better generalization performance and robustness than traditional methods.This thesis started with improving the feature utilization and feature expression ability of convolutional neural networks,and explored methods to improve the accuracy of facial expression recognition,then applied them to learning facial expression recognition.The main work of this thesis is as follows:(1)Aiming at the problem that the accuracy of the classic LeNet-5 network was not high on the expression dataset,an improved LeNet-5 network was designed.The expression recognition performance of the network was improved by changing the size and number of convolution kernels,increasing the number of network layers,and introducing Dropout mechanisms and L2 regularization.In addition,it also designed a cross-connection method that used 1×1 convolution while using different levels of features to adjust the proportion of different levels of features,which improved the utilization of features.Comparing the experimental results of the improved model and the classic model on the basic expression CK+ dataset and the learning expression OL-SFED dataset,it was proved that the improved model greatly improved the accuracy of expression recognition.On this basis,the influence of different data enhancement methods on the network performance was explored to further improve the effect of expression recognition.(2)In order to improve the feature expression ability of convolutional neural network,a dual channel attention module was proposed and a dual channel attention network was constructed.The module used convolution kernels of different scales to convolute on the two branches respectively.After connecting the feature maps of the two branches in the channel dimension,it used 1×1 convolution to reduce the dimension.It can obtain the features of different scales and reduce the number of parameters.At the same time,spatial attention mechanism and channel attention mechanism were introduced to improve the attention of important spatial features and channel features.Experiments on the two datasets showed that the dual channel attention network achieved good recognition results.(3)According to the learning facial expression recognition model finally obtained in this thesis,a learning facial expression recognition platform was built.The platform can effectively identify the five academic emotions of confusion,distraction,neutral,enjoyment,and fatigue that appear in the learning process by obtaining the facial expression information of students.The platform can also conduct an overall analysis of academic emotions in the learning process,and provide a certain reference for the judgment of students’ academic emotions in online education.