Research On Facial Expression Recognition Based On Convolutional Neural Networ

As a non-verbal communication carrier,facial expressions play a very important role in daily communication.With the ever-changing science and technology,the use of computers to recognize facial expressions has become one of the current research hotspots,and facial expression recognition technology is also widely used in intelligent security,intelligent driving,intelligent medical care and other fields.Although there are currently many facial expression recognition algorithms based on convolutional neural networks,the accuracy of facial expression recognition and the actual application effect are still not ideal.The main reasons are: 1)The difference between facial expression samples is small,and the difference within a class is large;2)loss of key expression features;3)Most facial expression recognition algorithms are in the research stage,ignoring the differences between facial expression images in experimental scenes and natural scenes,resulting in poor practical application results.In view of the above problems,this paper carried out the following research:(1)Aiming at the problem of small differences between classes and large differences within classes of expression samples.Improved on the basis of Res Net-18,a facial expression recognition method based on Diverse Branch Residual Network(DBRNet)is proposed.The network uses diverse branches to enrich the face feature space;uses the spatial attention residual module(SAR)to highlight the face features with discriminative significance.DBRNet was tested on the RAF-DB and CK+ datasets,and the accuracy rates reached 80.31% and 96.41%,respectively.The experimental results show that the network has a certain competitive advantage,which is conducive to improving the accuracy of facial expression recognition.(2)On the basis of further expanding the differences between classes and narrowing the differences within classes,a facial expression recognition method based on Multi-Branch Compact Bilinear Pooling Network(MCBP)is proposed to solve the key expression feature loss problem generated in the process of layer-by-layer convolution and pooling of convolutional neural networks.Based on DBRNet,the network incorporates Compact Bilinear Pooling(CBP)and Self-attention Mechanism(SA)to enhance the interaction and correlation between features.The multi-branch feature fusion strategy is proposed to avoid the loss of key expression features,expand the differences between expression classes,and narrow the differences within classes.MCBP was experimented on RAF-DB and CK+ datasets,and the accuracy of expression recognition was 84.58% and 98.46%,respectively.Experimental results show that MCBP effectively improves the accuracy of expression recognition.(3)In view of the difference between the experimental scene and the natural scene,the actual application effect of facial expression recognition is not ideal.A facial expression recognition system was developed using a We Chat applet.The experimental results show that the system has strong anti-interference ability,high recognition accuracy and strong practical application.