Capsule Model For Human Facial Expression Recognition Algorithm

In daily life,people mainly convey psychological and emotional information through facial expressions.Facial expression recognition is to use the constructed model to recognize facial expressions,and judge psychological emotions based on the expressions.Facial expressions recognition based on deep learning has received extensive attention and key research in the fields of human-computer interaction and medical treatment,and has also made great progress.The convolutional neural network in deep learning has good feature extraction capabilities for facial expressions,but it does not perform well in terms of spatial information extraction capabilities.In this thesis,we build a facial expression recognition model based on Capsule.The model is mainly divided into three parts: a 2-4 convolution layer that extracts basic features,a Capsule layer that represents and classifies entity features,and a decoder structure for further optimization of the network.The model in this thesis uses Capsule to represent facial expression entities,which can better extract spatial information in facial expression images.Aiming at the constructed Capsule facial expression recognition model,this thesis has done two aspects of optimization research.On the one hand,this thesis studies the influence of the convolutional layer on the final representation of the facial expression entity of Capsule.Due to the different features extracted by different convolutional layers,the original Capsule generated is also different,resulting in different recognition performance of the model on facial expression.On the other hand,this thesis studies the impact of the decoder on this model.This thesis studies two methods to construct a decoder for facial expression recognition,one is a decoder based on full connection,and the other is a decoder based on deconvolution.The decoder optimizes the model by reconstructing the image,and regularizes the model to prevent overfitting.In the model constructed in this thesis,the Re LU activation function is used in the convolution layer to speed up the training of the model.In order to verify the effectiveness of the model,experiments were conducted on the public dataset of facial expressions,and the results show that:1.Different convolutional layers have an influence on the experimental results,and the facial expression recognition model constructed by three convolutional layers has the best effect.2.From the experimental results of the model on the public data set,the decoder based on deconvolution is superior to the decoder based on full connection.3.The experimental results of the FERCaps-3-De model in this thesis perform better than the existing algorithms,and its accuracy rate reaches 98.2% on CK+ and 88.3% on JAFFE.