Research On Expression Recognition Technology Based On CNN Integrated Feature Fusion

Facial expressions are the external expression of personal emotions and the manifestation of complex psychological processes.Facial expression recognition,as an important branch of face recognition,has always been a hot spot in the field of artificial intelligence research and has broad application prospects,such as applications in remote education,safe driving,marketing assistance,criminal investigation and other fields.Due to the complexity and subtlety of expressions,real-time facial expression recognition is still a big problem.In order to solve this problem,this thesis proposes a facial expression recognition method based on convolutional neural network ensemble learning.At the same time,in order to solve the problem of uneven distribution of the data set,local features and overall features are fused.The main content of this thesis is as follows:First,select three classic convolutional neural networks,namely AlexNet,VGGNet,and ResNet.Modify the number of layers of these three networks to less than 10,and adopt the method of Stacking ensemble learning to integrate the three networks into a strong classifier,the last fully connected layer of the network is removed,and the SVM classification algorithm is used for the final classification and recognition of expressions.After setting the parameters,select the FER2013 data set as the training data set,train the three classic networks and the integrated convolutional neural network to convergence,compare and analyze the experimental results,and get the final recognition rate of 70.84%.Then,aiming at the problem of uneven expression recognition rate caused by uneven distribution of expression data,a method of fusing local features with overall features is used.Since the feature information of expressions is mainly related to the three organs of nose,eyes,and mouth,in order to remove redundant redundant information,the harr cutter is used to crop the expression pictures in the data set into three parts,namely nose,eyes,and mouth.Send these three local feature pictures to three modified convolutional neural networks for pre-training,and use the idea of ensemble learning to fuse these local features.Through the final confusion matrix graph,it can be seen that the recognition accuracy is improved to71.27%.Finally,the trained model is applied in practice.This thesis uses the PyQt5 library designed based on the Python interface to complete the design of the entire expression recognition system.The designed system has the following functions: Able to choose the model to realize the function of facial expression recognition;Able to recognize static facial expression pictures;Able to display the time of recognition;Able to turn on the camera to recognize real-time facial expressions.The test results show that the system can accurately recognize seven facial expressions and has a faster recognition speed.This thesis adopts ensemble learning algorithm,integrates three classic networks,and adopts related optimization algorithms such as expanded training data set to further improve the accuracy of facial expression recognition.The experiment was carried out on the FER2013 data set,and the experimental results show that the model has an expression recognition accuracy rate of 71.27% on the FER2013 data set.The final expression recognition system can realize real-time,high-performance expression recognition.The method can solve various problems caused by facial expression recognition errors,can be widely used in people’s daily life,and improve people’s work efficiency and quality of life.