Research And Implementation Of Facial Expression Recognition Method Based On Convolutional Neural Network

With the rapid development of computer technology and computer vision,the facial expression recognition algorithm plays an important role in the human-computer interaction and it has a very broad application prospect.The facial expression recognition algorithm can count the students'listening status in the classroom,record the driver's fatigue during driving,capture the abnormal expression of dangerous molecules in public places to avoid unnecessary accidents,and so on.Improving the accuracy of the facial expression recognition algorithm is an important task in the field of facial expression recognition.Choosing the appropriate network structure and loss function is critical to improve the accuracy of the facial expression recognition algorithm based on convolutional neural networks.The facial expression has the problem of large intra-class difference and high similarity between classes.The traditional convolution structures have a large number of parameters and a large amount of computation.Aiming at these problems,this paper proposes a loss function based on the cosine distance and the traditional loss function,which can reduce the intra-class feature differences in the feature space,increase the inter-class feature distribution,and improve the feature discriminating effect.At the same time,this paper designs a network structure based on deep separable convolution,which greatly reduces the parameter quantity and streamlines the network model.The facial expression recognition algorithm based on convolutional neural network designed in this paper adopts the structure of deep separable convolutional neural network,and uses the cosine distance loss function of the superparameters ? to take 0.01 and ?₁ to take 10 for supervised training.After a lot of experiments and analysis,the accuracy of facial expression recognition on the RAF-DB facial expression recognition dataset achieved 83.196%,which is better than the facial expression recognition algorithm supervised by the traditional loss function,and verifies the superiority of the proposed algorithm in facial expression recognition tasks.At the end of the thesis,the paper puts forward the construction process and implementation effect of the proposed algorithm in the facial expression recognition system.The system can recognize the facial expression in the video stream of the network camera in real time.For the face with certain deflection angle and occlusion,It has higher recognition accuracy and verifies the feasibility of the proposed algorithm.