The Research On Facial Expression Recognition Algorithm Based On Convolutional Neural Network

In recent years,with the continuous development of artificial intelligence,the field of image recognition has made great progress,and facial expression recognition has gradually become a hot research topic.It is widely used in intelligently monitor driving,remote network education and paramedical cares to analysis and recognize other individual's facial expressions by extracting facial expression features.Traditional facial expression recognition methods have disadvantages,such as complicated process with handling expression's feature,weak generalization ability,and poor applicability.Nevertheless,with the rise of deep learning,the rapid development of convolutional neural networks,facial expression recognition technique has a great improvement.This thesis does research on facial expression recognition method based on convolutional neural network.Aiming at the shortcomings of general convolutional models of weak recognition and poor feature extraction ability,this thesis builds a improved convolutional neural network(ContinuousNet model),based on the classic AlexNet network and refers to the VGG model.This model structure uses a continuous small convolution structure to replace the single large convolution structure,which not only enhances the model's ability to extract features but also enhances its linear expression ability.Small convolution kernel and step size are adopted in parameter setting,which makes it easier to collect local difference information of images.This method will use the CK+ expression database and the FER2013 expression library for simulation and compare experiments with the AlexNet model.The experimental results show that the recognition accuracy of ContinuousNet model is improved compared with AlexNet model.Therefore,the ContinuousNet model proposed in this thesis is better than AlexNet model in facial expression recognition.In order to solve the problem that the neural network model will lose some low-level detail features in the convolution transform process,this thesis based on the ContinuousNet model of the improved convolutional neural network and the structure of the RESNET model,designs a cross-layer feature fusion convolutional neural network(SFCNN model).Firstly,facial expression features are extracted by ContinuousNet model.Secondly,the output features of different levels are fused to improve the utilization rate of features and avoid the loss of detailed features.Principal component analysis(PCA)is used to reduce dimension,eliminate feature redundancy and reduce the number of parameters.Finally,the processed fusion features are input to the full connection layer as the final classification features.In the amplified FER2013 dataset,SFCNN models with three cross-layer modes were compared with ContinuousNet models without cross-layer feature fusion.The results show that the recognition rate of SFCNN model using the cross-layer feature fusion method is higher than that of ContinuousNet model,which proves the effectiveness of the cross-layer feature fusion method.At the same time,three different cross-layer feature fusion methods were compared,and the results show that the SFCNN-3 model with three layers of feature fusion has the highest recognition rate.