Facial Expression Recognition Based On Deep Learning

In recent years,deep learning has become more and more concerned by people,and deep learning has been continuously developed.It has already appeared in different fields,and deep convolutional networks,as one of deep learning,have enough data through convolutional layers,pooling layers,etc.A highly efficient deep convolutional network can be trained,especially in terms of image processing and recognition.Deep convolutional networks are now even unparalleled.Facial expression recognition based on deep learning is based on the wide application and high practical value of facial expression recognition at this stage.This paper proposes three facial expression recognition algorithms based on deep learning facial expression recognition.The first is a single-channel facial expression recognition based on a deep convolutional network.The convolutional layer and the pooled layer used in the network structure have 8 layers.The convolutional layer and the pooled layer are followed by the fully connected layer and the output of softmax.Floor.At the same time,the deep convolutional network structure adopts standardization and Dropout,which avoids the over-fitting problem that is very common in deep learning.When training deep convolutional networks,Adam's adaptive learning rate algorithm is adopted.It is very helpful to set the learning rate for each part of the network through Adam.Based on the single-channel network structure,facial expression recognition based on local binary mode and deep convolutional network is proposed.The network structure is a structure consisting of three layers of convolutional layers and two layers of pooling layers.The training and recognition process is to first standardize the picture and process the local binary mode,and then send the processed picture to the convolutional network structure.Because the local binary mode is processed,the next layer of the network is more It is easy to learn from the network of the previous layer,which reduces the level of network structure,and because of the reduction of the network level,the gradient problem is also solved.This paper also proposes a two-channel network structure for facial expression recognition.Specifically,one channel is composed of 8 layers of convolutional layer and pooled layer,and the other channel is composed of 5 layers of convolutional layer and pooling layer.There are fully connected layers after the two channels,and finally the fusion layer and the output layer.This paper has trained and tested on the three face expression databases JAFFE,CK+ and FER2013.Compared with some traditional image processing algorithms and machine learning,the recognition rate has made great progress,and the algorithm proposed in this paper is verified.Effectiveness.