Research On Human Face Expression Recognition And Application Based On Deep Learning

With the rapid development of artificial intelligence,traditional facial expression recognition technology is difficult to meet more precise human-computer interaction needs.Research on facial expression recognition based on deep learning has received great attention from researchers at home and abroad.However,there is still room for optimization in the structure and parameters of deep convolution learning networks for facial expression recognition.Therefore,it is important to study deep convolutional neural networks with more optimized features for improving facial expression recognition,because it is full of practical application value.The main research contents of the thesis are as follows:(1)In view of the fact that the detailed network design of the existing deep learning method in expression recognition is not optimized enough,it is easy to cause poor test results and the single convolutional neural network model can not combine the defects of expression sequence data,this paper designs a depth structure based on expression sequence,Deep Convolutional Neural Network Based on Expression Alignment(DCNNBEA).It is based on the VGG16 convolutional neural network,the structure is designed to adjust the number of layers of convolutional neural networks,extract facial expression features efficiently and accurately,and combine long-term short-term memory(LSTM).LSTM can do sequence prediction analysis.Finally compared with other similar methods on the fer2013 data set,the experiment obtained relatively good results.(2)For the face expression recognition based on depth learning algorithm,the gradient disappearance and the parameter initialization of the deep learning model are easy to occur during the training process,so a trainable Multi-layer Network Activation Function(TAFMN)is proposed in this paper The activation function has a powerful function mapping representation that extracts potential features by adding linear and nonlinear activation function.In addition,the relationship between the size and number of convolution kernels is studied,and the relationship between the optimal size and quantity parameters of the convolution kernel is found.Finally,the entire network parameters are integrated,and then the training and testing are also performed on the fer2013 dataset.Compared with the similar methods,the network method after optimizing the parameters obtains better results.(3)A human-computer interaction system based on facial expression recognition was designed and implemented.The system is mainly used to detect the effectiveness of the facial expression recognition model proposed in this paper,and it can also be applied to the human-computer interaction experience area of the mall,which can provide a good possibility for human-computer interaction behavior.Image processing is performed on the obtained facial expression image of the user,and then transmitted into the relevant trained facial recognition model,and finally the detection result is quickly obtained,and the result is combined with friendly,visualv text and charts.These can take good experience to users.