Research On Human Expression Recognition Method Based On Deep Learning

With the development of artificial intelligence technology and big data,machines begin to have the learning ability like the human brain and process huge information data.There is a lot of information in human facial expressions,and humans can understand each other's emotions and states through their expressions.And now through the method of intelligent image recognition,the computer can also achieve the acquisition and analysis of facial expression information.Facial expression recognition can be applied in many fields such as human-computer interaction,security,robot manufacturing,medical treatment,communications,and automobiles,and has strong market demand and broad development space.Therefore,the research of facial expression recognition and the improvement of recognition performance are of great significance.Artificial neural network is a mechanism built on the basis of biological neural network.Convolutional neural network is developed on the basis of artificial neural network,which can automatically extract features,avoiding the influence of artificially designed feature extraction scheme on the recognition effect.With the help of deep learning and deep VGG convolutional neural network,this paper builds a deep convolutional neural network model suitable for facial expression recognition.For the deep convolutional neural network expression recognition model training process,due to the problem of gradient disappearance and gradient explosion caused by the deepening of the network layer,the LReLU function is used as the activation function,which makes the gradient update steadily,and avoids the neurons that appear using the ReL function The problem of suppression.For deeper convolutional neural networks,which have stronger learning ability and are prone to overfitting,cross-entropy function is used as the target function,and L2 regularization is performed on the target function to alleviate the phenomenon of overfitting.Particle Swarm Optimization(PSO)algorithm is a swarm intelligence optimization algorithm,which has the characteristics of easy implementation and fast convergence speed.In order to improve the accuracy of expression recognition and the convergence speed of the model,this paper uses the PSO algorithm to optimize the initial parameters of the deep convolutional neural network.This method can not only improve the convergence speed of model training,but also avoid the randomness of taking initial parameters from normal distribution.And the PSO algorithm does not depend on the gradient when updating the position,which can avoid the problem of gradient reduction or even disappear due to the initialization method.Although thePSO algorithm has a fast optimization speed,it is easy to fall into the local optimal solution.In order to solve the problem of poor global search ability of PSO algorithm,this paper integrates the cross mutation method in genetic algorithm(GA)into PSO algorithm,and uses the fused GA-PSO algorithm to optimize the initial parameters of deep convolutional neural network,making the optimization effect Better,further improving the accuracy of facial expression recognition.This paper uses Fer2013 expression dataset and CK + expression dataset to experiment and verify the above methods respectively.In the end,the expression recognition model constructed in this paper achieved an accuracy rate of 71.93% on the Fer2013 expression data set and 96.7%accuracy on the CK + expression data set,and the experimental results showed that the method proposed in this paper made the model convergence rate certain promotion.