Research On Facial Expression Recognition Method Based On Convolutional Neural Network

With the continuous development of intelligent computing and deep learning,facial expression recognition has become one of the hot topics in the field of computer vision.It has a wide range of application prospects in human-computer interaction,intelligent driving,intelligent medicine and so on.The methods of facial expression recognition mainly include traditional machine learning method and convolution neural network method in deep learning.With the rapid development of deep learning,more and more researchers begin to study facial expression recognition through deep learning algorithm.In this paper,the difficulty of facial expression recognition is analyzed based on deep learning algorithm.The main work is as follows:(1)In view of the poor robustness of traditional machine learning methods in extracting facial expression features and the problem that existing convolutional neural networks cannot fully extract facial expression features,In this paper,a convolutional neural network method combining attention mechanism is proposed.First,RESNET is used as the basic model of the network to avoid the problems of gradient disappearance and the decrease of accuracy due to the deepening of network layers.Then,the attention mechanism module of CBAM is introduced to generate the attention map in the channel domain and the spatial domain of the feature map,after multiplying with the original feature map,a recalibrated feature map is obtained.In the training process,the improved loss function A-softmax is used to generate the angle interval by manipulating the feature surface,so that the different types of features it learns always have the angle interval in the angle space,and the similar features gather more closely.The recognition rate is 1.1% and 1.3% higher than that of Res Net by verification on FER2013 and JAFFE data sets respectively.Experimental results show that this method effectively improves the ability of network feature expression,enhances the ability of distinguishing different facial expression features,and achieves good recognition performance.(2)Aiming at the problem that the common real-time face detection algorithm MTCNN is not accurate enough in the complex environment and the detection speed is low,an improved mtcnn algorithm MTCNN-DP-FPN is proposed.Firstly,the depth separable convolution is used instead of the conventional convolution in MTCNN to reduce the network parameters and calculation cost,and the feature pyramid network(FPN)is added to the convolution layer to make the output features have high resolution and high semantics.The MTCNN network and MTCNN-DP-FPN network are trained respectively on the WIDER FACE database of face detection,and the comparison experiments are carried out on the FDDB face database.The experimental results show that the accuracy of face detection is 95.5%,and the detection speed is 160 fps,which can meet the requirements of real-time face detection.Finally,this paper designs a real-time facial expression recognition system based on facial expression recognition algorithm and face detection algorithm,and shows the effect of the system through the testers.