| In daily communication,the facial expression is an important way for humans to express emotions and convey information.With the rapid development of artificial intelligence,it is an inevitable trend to make machines understand people’s feelings in future.As an important part of affective computing,facial expression recognition has a wide range of applications in education,human-computer interaction,medical treatment and other fields.Decause the facial expression changes slightly between different expressions,it is challenging for a computer to accurately recognize facial expressions.This dissertation proposes two models which use attention mechanism and convolutional neural network.The main work of this dissertation is as follows:Firstly,this dissertation proposes an expression recognition model based on the fusion of the attention module and the lightweight convolutional neural network Shuffle Net V2.Firstly,this dissertation researchs the two attention modules,SENet and CBAM,how to enhance the performance of convolutional neural networks.And then this dissertation determines the network structure of the lightweight convolutional neural network Shuffle Net V2.Finally this dissertation fuses the attention module and the Shuffle Net V2 model to obtain the fusion model I and the fusion model II.The optimization algorithm is Adam algorithm,and the cross entropy function is selected as the loss function.Secondly,this dissertation proposes an expression recognition model based on fusion of the attention module and Dense Net.Firstly,this dissertation explores the network structure of Dense Net.And this dissertation fuses respectively the attention module,SENet and CBAM,and Dense Net model.Finally this dissertation can obtain the fusion model Ⅲ and the fusion model Ⅳ.In the process of training the model,the Adam algorithm is used for optimization,and the cross entropy function is used as the loss function.Finally,the model is trained on three data sets of CK+,FER2013,and Exp W,and the recognition results are obtained and analyzed.Firstly,the dissertation uses the MTCNN algorithm to detect human faces.After the online data enhancement method is used to expand the data set,models are trained in three data sets.The final experimental results show that the recognition effect of the fusion model has been improved. |
