Research On Facial Expression Recognition Based On Small-scale Kernel Convolution

As a main form of conveying emotional states and intentions,facial expressions provide important nonverbal communication cues in interpersonal relationships.With the rise of technologies such as artificial intelligence and machine learning,the existing human-computer interaction methods have been unable to meet human needs,giving machines the ability to accurately recognize human emotions has gradually become a hot research issue in the fields of computer vision.Affected by the objective conditions such as illumination and angle,it is difficult to achieve the unity of facial expressions,which poses a huge challenge for the computer to recognize emotions.Therefore,building a machine model with accurate facial expression classification ability and learning different facial expression features is of great research significance for realizing human-computer interaction.With the wide application of deep learning methods in the field of computer vision,the effect of existing facial expression recognition models has been significantly improved.However,there are still problems such as the generalization ability of network is weak,the complicated structure of the model leads to a large amount of network computation,and expression recognition result is insufficient.In order to effectively improve these problems,this paper proposes a facial expression recognition method based on small-scale kernel convolution.By designing multiple layers of small-scale kernel convolution blocks to extract facial expression features,training and testing the sample data to optimize the feature extraction network.The model uses the Softmax classifier to recognize the facial expression,and the performance of the method is experimentally verified to prove the effectiveness of the algorithm.The research work of this paper mainly includes the following aspects:1)Using data enhancement transformation and other methods to enhance the generalization ability of the network model and reduce the diversified impact of facial expression images.In the image preprocessing stage,face detection and data enhancement are used to expand the sample data.In the testing phase,data enhancement transformation is used to increase the robustness of the model.2)In order to effectively extract facial expression features while reducing the complexity of the algorithm,this paper design a small-scale kernel convolution block structure.Multi-layer small-scale kernel convolution is used to replace the large convolution function to ensure that sufficient receptive field size is obtained while reducing model network parameters.The network uses a fully connected layer to integrate feature information before outputting the classification,ensuring the simplicity of the algorithm.3)Construct a facial expression recognition model based on small-scale kernel convolution,iteratively learn image expression features through multi-layer small-scale kernel convolution blocks and down-sampling network structure.Applying Batch Normalization and Leaky-ReLU activation function to improve the model's nonlinear capability,and adding the Dropout method to the fully connected layer to reduce the effect of overfitting.The Softmax classifier is used to recognize the seven kinds of facial expression,and the cross-entropy loss function is introduced to optimize the facial expression extraction ability of the model through back propagation to achieve accurate recognition of the target facial image expression.In this paper,we use two public data sets,FER2013 data set and CK+data set.The experimental results show that the facial expression recognition method designed in this paper can effectively recognize facial expressions,reduce the algorithm complexity,and improve the accuracy of facial expression recognition.