Research On Expression Recognition Method Based On Convolutional Neural Network

Facial expressions are intuitive reflections of human psychological state,contain rich emotional information,and are one of the important ways of communication between people.With the rapid development of artificial intelligence and computer technology,facial expression recognition technology has become a current research hotspot,and has a wide range of applications in education,medicine,traffic safety,human-computer interaction,psychology and other fields.Due to the complexity and variability of human facial expression features,traditional facial expression recognition technology has the disadvantages of insufficient feature extraction and susceptibility to external environmental influences.The convolutional neural network can automatically extract deeper essential expression features from a large amount of data,reduce the influence of human factors and the external environment,make it have strong generalization and improve the final recognition accuracy.With the goal of further improving the accuracy and robustness of facial expression recognition algorithms,this article conducts a series of exploration and research on facial expression recognition methods based on convolutional neural networks,and builds a facial expression recognition system The main research contents of this article are as follows:(1)Aiming at the problems of large amount of convolutional neural network model parameters,slow training speed and inability to extract key features of expressions,a lightweight expression recognition method with efficient channel attention network is proposed.This method is based on the depth separable convolution to achieve linear bottleneck structure improvement,reduce network complexity and prevent over-fitting.By designing an efficient attention module,the depth of the feature map is combined with spatial information,focusing more on the extraction of important features,and adopting joint The loss function makes the network have a better feature discrimination effect,reduces the feature difference within the same expression,expands the feature distance between different expression classes,and ensures the accuracy of classification.And the recognition performance of this method is verified on the FER-2013 and CK+data sets.(2)Aiming at the problems of the existing artificially annotated facial expression data sets still having annotating uncertainty,and the convolutional neural network extraction features are not perfect,a facial expression recognition method based on selfcure and multi-feature fusion improved VGGNet is proposed.First,add a multi-scale feature extraction network structure at the front end of the VGGNet16 network,and perform branch feature fusion with multiple features extracted from networks of different depths to retain more complete feature information.Then,by using batch normalization after each convolutional layer and using Dropout after the fully connected layer,the convergence rate of the network model is accelerated,and the probability of gradient disappearance and overfitting of the network model is reduced.Finally,a self-correcting network is added to the back end of the network to suppress the influence of the labeling uncertainty of the sample on the network model training,and the superiority of the proposed method is demonstrated on the FER-2013 and RAFDB data sets.(3)Design a set of facial expression recognition system.The system performs facial expression recognition based on the model trained by the facial expression recognition method proposed in this paper,and experimental verification is carried out on static facial images and real-time dynamic videos.The system not only achieves high recognition accuracy,but also satisfies real-time facial expression recognition.Sexual requirements.