Face Expression Recognition Based On Convolutional Neural Network

Facial expressions are an important medium for people to convey their inner emotions,which not only plays a role that cannot be ignored in interpersonal communication,but also an important part of human-computer interaction.In recent years,with the update and development of computer information technology,face expression recognition technology has become increasingly mature and has played an important role and significance in many fields such as public safety,safe driving,assisted medical treatment and recommendation systems.In this era of the integration of the Internet into all walks of life,in order to quickly and accurately identify user expressions,face expression recognition technology is becoming more and more important.The traditional face expression recognition method based on machine learning relies on the knowledge and experience of the designer,which is not only difficult,time-consuming and laborious to design,but also causes certain losses to the original feature information,and cannot ensure the optimality of these features.As a result,deep learning,which can learn from data and extract features on its own,is gradually coming into people’s sights.In view of the superior performance of convolutional neural networks in deep learning in the field of image recognition,more and more research on face expression recognition technology is carried out on convolutional neural networks.In the process of research,various deficiencies and problems of facial expression recognition technology have gradually been exposed,and this article will study and carry out work on some of them.Aiming at the problems of gradient disappearance and network degradation with the deepening of convolutional neural networks,the problems of insufficient sample size and uneven distribution of sample categories in face expression datasets,and the problems of background interference and redundant information affecting the speed and accuracy of model training faced by face expression recognition in practical applications.In this paper,an improved Res Net50 network model based on coordinate attention mechanism is proposed.The coordinate attention mechanism module is selected and embedded in the Res Net50 network model to focus on the key features of facial expressions,alleviate the problem of information overload,and enhance the representation ability and robustness of the model.The Adam optimizer combined with exponential decay learning rate is used to optimize the hyperparameters of the improved network model,which further improves the model training efficiency and recognition accuracy.Database preprocessing methods such as face clipping and data enhancement and weighting operations on cross-entropy loss function were used to enrich the face expression dataset and reduce the impact of uneven distribution of sample categories.Aiming at the problems inherent in traditional convolutional neural network frameworks,long-distance dependencies between features and redundancy between internal channels,and some expressions are similar,which reduces the accuracy of model recognition.In this paper,a Res Net50 network model based on Involution operators and hierarchical bilinear pooling is proposed.The Involution operator is introduced to replace part of the convolution operation in the convolutional neural network to expand the spatial information interaction area of the image,improve the model’s ability to extract long-distance dependent feature relationships,reduce model parameters and computation,improve the efficiency of the model,and enhance the model’s attention and robustness to the key information of facial expressions.Hierarchical bilinear pooling is adopted to perform feature fusion to enhance the representation ability of features and strengthen the ability of the model to distinguish subtle differences between expressions.A Dropout layer is added to prevent overfitting and improve the generalization ability of the model.