Research On Facial Expression Recognition Based On Convolutional Neural Network

Traditional facial expression recognition feature extraction and expression classification are completed separately.It is necessary to manually design the algorithm for extracting expression features.To extract good features,it takes a lot of time to design and extract representative facial expressions,Convolution Neural network(CNN)proposes that this working mode is broken.Unlike traditional facial expression recognition,CNN does not need to manually extract features and then classify it.It can use the entire picture as a model.Input and output the result of facial expression classification after a series of operations.This process is like a black-box operation process.The classification result can be obtained without knowing what each step does.The significant advantage of CNN in the field of image recognition is that it can extract high-quality features without human intervention.However,when traditional CNN is used for facial expression recognition,only the features extracted from the last top layer are used and then input into the classifier for classification.Using the problem of incomplete facial expression features,facial expression recognition is essentially an image classification problem,so there will be problems in the recognition process such as a large amount of parameters,a large amount of calculation,and a long timeconsuming.At the same time,the existing facial expression recognition is used for facial expression recognition.Most of the methods have disadvantages such as low recognition rate and weak generalization ability.According to the above questions,this paper combines CNN network model,multiscale pooling,feature fusion and extreme learning machine together to form a method for facial expression recognition.The method first uses convolutional neural network to extract multiple faces Expression feature maps.Secondly,the last three layers of feature maps extracted by CNN are used to obtain three feature vectors by multi-scale pooling operation.Then,the three feature vectors are cascaded and merged into a facial expression feature vector,which has The nature of multi-scale and multi-attribute can express facial expression features well.Finally,the facial expression feature vector with multi-scale and multiattribute after cascade fusion is used as the input of the extreme learning machine classifier to train it,and finally by the classification The device outputs the classification result of the expression classification.The experimental results show that this method can effectively improve the accuracy of facial expression recognition and shorten the training time of the model.At the same time,the design experiment verifies that the model has strong generalization ability.Figure 41;Table 6;Reference 52...