Research On Facial Expression Recognition Algorithm Based On Convolutional Neural Network

In the current facial expression recognition task,Convolutional Neural Networks(CNN)and Local Binary Patterns(LBP)can only extract single features of facial expression images during facial expression feature extraction,which makes it difficult to extract precise features highly related to facial changes,thus affecting the effect of facial expression recognition.In addition,the traditional loss function is difficult to distinguish the inter-class distance and intra-class distance in facial expression features,so it is impossible to distinguish the extracted features effectively.Therefore,the in-depth study in this paper is based on the simplified VGG convolutional neural network as the framework,combined with the new additive angle margin loss function,which is of great significance for improving the accuracy of facial expression recognition.This paper proposes a facial expression recognition method using feature fusion based on VGG-NET.The method combines the LBP feature and the features extracted by the CNN convolutional layer into the improved VGG-16 network connection layer by weighting.Finally,the fusion feature is sent to the Softmax classifier to obtain the probability of various features,and complete the basic six expression classifications.The experimental results show that the average recognition accuracy of the proposed method on the CK+ and JAFFE datasets is 97.5% and 97.62%,respectively.The recognition results obtained by the fusion features are significantly superior to that of single feature recognition.Compared with other methods,this method can effectively improve the accuracy of expression recognition and is more robust to illumination changes.This paper also proposes an improved additive angular margin loss function,which improves the traditional Softmax loss function by increasing the margin m between the angle of the feature and the target weight and the non-target weight to reduce the difference of intra-class features and increase the distribution of inter-class features,thereby improving the effect of feature discrimination.Experimental results show that the average facial expression recognition accuracy of the proposed method on the CK+ and JAFFE test sets is 98.87% and 98.92%,respectively.The effect is superior to the facial expression recognition model trained by the Softmax loss function,the additive margin softmax loss function and the angular softmax loss function.