Study On Facial Expression Recognition Algorithm Based On Deep Learning

Facial expression is the most intuitive way to transmit human emotional state.Through analyzing facial expression,one can obtain the mental and physical condition of someone at a certain time.Therefore,facial expression recognition has important application value in the fields of human-computer interaction,automatic driving,medical treatment,and has become a hot research topic.With the popularization of deep learning,the research on facial expression recognition has gradually changed from traditional image processing to deep learning.However,in real life,the number of samples of the exact expression data set is small,and affected by hardware equipment,which leads to the limitation of the method of increasing the depth of network model to improve the accuracy of expression recognition.This paper studies the expression recognition algorithm,and the main work is as follows:(1)Study on facial expression recognition based on improved AlexNet.In order to solve the problem that AlexNet has low recognition rate in facial expression recognition,on the basis of deep study on AlexNet network,inspired by VGG network architecture,the AlexNet is optimized.The convolution kernel of 11×11 is replaced by the convolution kernel of 5×5,and the cascaded small convolution kernel is used to replace the large convolution kernel of 5×5.The asymmetric convolution is introduced into the specific layer of the network to increase the nonlinear mapping and reduce the parameters of the model.Then the batch normalization operation is used to accelerate the convergence speed of the network.Compared with the original network,the improved network has better feature extraction ability.(2)Study on facial expression recognition algorithm based on dual-branch feature fusion.To solve the problem that the feature extracted by convolution neural network ignores the subtle changes in the active region of facial expression,the expression recognition algorithm based on dual-branch feature fusion is designed.In the first branch,Gabor feature of ROI region is used as input.In order to make full use of the detailed features of the active facial expression region,the active facial expression region is segmented from the original face image,and Gabor transform is used to extract the feature of the region,the texture feature focuses on the detail description of the local region.In the second branch,the complete pure face image is used as input,and the improved AlexNet network is used to extract feature,the feature is intended to represent the integrity of the expression,and the two features are complementary.Finally,the fused features are used for expression classification.The simulation results on different datasets show that the algorithm has a better recognition result.(3)Study on face expression recognition based on the variational autoencoder.The problem of difficulty in network training due to fewer expression data sets,which affects the accuracy of expression recognition.so the variational autoencoder is applied to facial expression recognition,and the attention mechanism is embedded in the network of variational autoencoder,which can make the network pay more attention to the useful features.The variational autoencoder is trained by unsupervised learning,the encoder of the trained variational autoencoder can well extract features from the input image,so the encoder is used as a feature extractor.The full connection layer and softmax layer are added after the feature extractor.Then the labeled dataset is used to fine tune the network in a supervised way to give it the ability to distinguish different facial expressions.The simulation results on different datasets show that the algorithm has a better recognition result.