Research On Facial Expression Recognition Based On Deep Learning

With the rapid development of artificial intelligence and computer vision,human-computer interaction has attracted extensive attention.Because the difference between different human expressions is very small,it is a very challenging task for computers to accurately recognize facial expressions.Face Expression Recognition(FER)is to use computer to analyze human facial expressions and then recognize the real emotions of the face,which has practical application value for human-computer interaction.Traditional facial expression recognition methods are based on the features manually designed by researchers,which do not require high computer performance,but the recognition rate is limited.In order to solve the above problems,this paper adopts deep learning method to recognize facial expressions.In this paper,facial expression recognition is mainly carried out on static pictures.The main research work is as follows:1)The face image is clipped,then the clipped expression image is processed by scale normalization based on bilinear interpolation and gray normalization based on histogram equalization algorithm,and finally the facial expression image with uniform scale and gray is obtained.2)Aiming at the shortcoming of difficult convergence of loss function in expression recognition,double loss function is used to supervise the training of network model by combining center loss and cross entropy loss,so as to ensure correct classification,reduce the intra-class gap of each expression,increase the distance between classes,and further improve the recognition rate of expression.3)An improved facial expression recognition method based on cross-connection convolution neural network is proposed.On the basis of convolution neural network,the pooling layer in the network is cross-connected,then the fused features are sent to the full connection layer,and finally the Softmax classifier is used to classify them.The use of the activation function SeLU in the network will accelerate the convergence speed of the network in the training process.The improved convolution neural network model method is used to carry out facial expression recognition experiments on FER2013 and CK+expression databases respectively.Compared with other convolution neural network experimental results,the effectiveness of the proposed method is verified.The experimental results show that the improved cross-link convolution neuralnetwork facial expression feature extraction method proposed in this paper can recognize facial expressions more accurately and has strong generalization ability.