Facial Expression Recognition Based On Sparse Autoencoder And Attention Mechanism

With the rapid development of science and technology,the technology of recognizing subtle facial expressions is continuously increasing.It can better recognize the subtle differences of human expressions in real life,accurately detect human emotions,psychological conditions and intentions,and is expected to be widely used in human-computer interaction,smart transportation,distance education and so on.Traditional face recognition technology is based on machine learning algorithm to extract image features and then select a classifier for classification.This technology will lose some important information when extracting the feature values,and cannot extract more important feature information in a specific way,which will have a certain impact on the final recognition results.Therefore,this paper chooses deep learning technology to recognize facial expressions,and after analyzing the related technologies of deep learning,chooses the appropriate network framework to improve.The main work of this paper includes the following parts:(1)In order to solve the problem that the model cannot extract sufficient features from images in shallow network structure,a hybrid neural network structure is proposed.The network includes sparse autoencoder and convolutional neural network(SCNN).The network first reconstructs the original input image by a sparse autoencoder to learn the approximation between the original data and the reconstructed data,to get more high-dimensional and abstract features,and then further extract the features and reduce the dimensionality.by combining convolution neural network.(2)In order to extract important features from images,this paper proposes a facial expression recognition model based on attention mechanism and convolutional neural network.The model combines channel attention and spatial attention,which can quickly find the focus of attention in the image,extract important features in the image according to the weight of information,weaken useless information,improve the efficiency of network operation,and improve the shortcomings of traditional convolutional neural network.(3)In order to verify the effectiveness of the model,this experiment uses CK + and FER2013 to Cross-validation the model.In the training process,this paper first input the CK+ database into the model for train,and then the trained model is put into the FER2013 dataset for verification.Experimental results show that the two models proposed in this paper have better efficiency and robustness than other models,and can achieve better results in multiple data sets.