Research On Facial Expression Recongnition Based On Deep Convolutional Neural Network

In recent years,with the rapid development of Internet technology and computer hardware,facial expression recognition technology has been applied in many fields,such as human-computer interaction and driver fatigue monitoring.Facial expression recognition is an interdisciplinary research field.In practical applications,there are many complex scenes that will affect the accuracy of recognition,such as rotation,offset,and lighting.Because many traditional identification methods are difficult to meet the actual needs.Some excellent traditional methods all need to manually set some information about the characteristics of the special zone.It is easy to cause imperfect consideration,resulting in information loss,and it is difficult to accurately identify.The convolutional neural network has great advantages in the processing of image data.Moreover,for the subtle changes of facial expressions,the convolutional neural network can continuously learn and update facial features to extract many deeper features.Based on the facial expression recognition algorithm,this thesis further studies the application of facial expression recognition in intelligent assisted driving.Therefore,for the above problems,this thesis mainly proposes a facial expression recognition algorithm based on deep convolutional neural network.The main research contents of this article are as follows:1.In view of the high variability of the subject in the face detection process,and the complex influencing factors and lighting factors in the driving environment,the traditional face detection method is difficult to achieve fast and accurate face detection and positioning.This thesis designs a three-layer cascaded convolutional neural network for face detection and localization.Through convolutional neural network with three different structures,the first stage roughly locates the face position through the shallow network,the second stage further refines the data of the first stage,and finally the third stage removes all non-face data,The final accurate location of the face can avoid the problem of low recognition rate caused by factors such as complex environment.In this thesis,through FDDB face data verification,we can see that the accuracy of face detection has been improved to a certain extent,and it has certain robustness.2.For the data-driven learning algorithm,the data set is the top priority.For the dataset of fatigue expression,Yaw DD and CEW are commonly used,but the dataset is only collected for a single state and the data set is constructed,and the data set is large Most of them are European and American faces,which do not meet the requirements of this article for the simultaneous detection of yawning and dozing.Therefore,this thesis trains the feature extraction network by simulating the real driving environment and selfcollecting data sets.3.Different expressions of different faces will show huge differences.The feature extraction of the existing learning algorithms is based on the manually set features,which loses many detailed features of the original image and fails to achieve efficient and realtime facial expression classification.Therefore,a deep convolutional neural network with multi-size convolution is designed in this thesis.Through multi-size convolution,different feature scales can be satisified with different needs,the problem of inadequate manual setting of feature points can be avoided.And in order to avoid the problem that the model runs slower Caused by too deep network,the parameter compression technology of neural network is studied in this thesis.Finally,global average pooling replaces the traditional full connection layer,and the number of parameters is greatly reduced without affecting the accuracy Experimental results show that the algorithm can improve the recognition accuracy and running speed to a certain extent.