The Research Of Facial Expression Recognition Based On Improved Deep Learning

Convolution neural network belongs to deep learning network architecture,its use local Shared awareness,parameters and operation parameters optimization,that need to be trained to reduce weight parameters,therefore,positive neural network can be in more complex environment,still has the good performance of feature extraction.At the same time,the convolutional neural network in the deep learning model has certain characteristics invariance for its own spatial structure,such as migration,scaling,etc.Automatic encoder is a kind of artificial neural network architecture,it is a special perceptron,for unsupervised learning.This paper USES the semi-supervised deep learning algorithm based on improved convolution neural network to study facial expression recognition.The main research contents are as follows:1.Method of extracting expression features of semi-supervised deep learning.In view of the learning method of convolutional neural network,supervised learning methods are generally adopted,but there are insufficient training samples in supervised learning mode and the problem of gradient dispersion.This article will for supervised learning and unsupervised learning algorithm,studies the convolution of the neural network based on automatic coding a semi-supervised deep learning model in feature extraction,feature extraction part and supervision and deep learning model in the convolution,the loss function of neural network used for the classification of cross entropy loss function.Softmax classifier is used to verify the expression feature extraction method of semi-supervised deep learning.The method is applied to face expression recognition of FER3013 data set.2.Regularized expression classifier.The classifier plays a very important role in the expression recognition system,and it is necessary to improve the robustness of classifier.In order to satisfy the requirement of system data on sparsity,the classifier adopts sparse representation.In order to improve the robustness of classifier,based on sparse representation classification,this paper studied the regularization coding classifier,enhance the robustness of classifier and improve the depth of a semi-supervised learning algorithm is applied to the recognition of facial expression recognition system.3.Expression recognition based on improved semi-supervised deep learning.Over fitting phenomenon existed in the work of deep learning problems,based on the feature extraction of supervision and deep learning model was improved,on the basis of the convolution in the depth of a semi-supervised learning framework on the loss function of neural network is introduced into regularization item,form the improved depth of a semi-supervised learning model.The feature extraction part and classification part of the recognition system are introduced regularization,which improves the over-fitting phenomenon of feature extraction and improves the accuracy of the recognition system.The study shows that regulatation has a very important influence on the performance of the recognition system,and performs the performance test on the FER2013 data set.The experimental results show that the improved depth learning algorithm is better than the semi-supervised deep learning model without regularization.