Children's Smile Recognition Based On Deep Convolutional Neural Network

In recent years,with the acceleration of urbanization and the implementation of compulsory education,the pressure of students' study and life is increasing day by day,and the occurrence of psychological behavior problems is becoming more and more prominent.Therefore,the school has taken the mental health education of students as the top priority in the teaching work.Effective expression classification can assist psychology researchers to study psychology and other disciplines.By classifying expressions and analyzing children's psychological activities and mental states,they can reduce the occurrence of psychological behavior problems.The smiling face,as one of the most common facial expressions of human beings,reflects the psychological state of human beings and conveys a wealth of emotion and intention information.In school,by encouraging students to smile more,so that students keep a happy and cheerful mood,which can effectively achieve students' happy growth and education,so accurately identify children's smiling face to help solve students' mental health problems is of great significance,there is a wide range of application prospects.Based on the deep neural network,this thesis conducts a study on the recognition of children's smiling faces.The main research work is as follows:(1)The database of children's smiling faces was established and preprocessed.Firstly,children's face pictures were collected through Internet search,and then some children's face pictures were generated by StyleGAN.A database of 9712 children's smiling faces was constructed,which was classified into 4615 laughing faces,2101 smiling faces and 2996 non-smiling faces through the scoring mechanism of multiple people.At the same time,aiming at the unbalanced problem of three types of data samples,this thesis used data enhancement technology to rotate and cut the pictures,and expanded the database of children's smiling faces into 20356 pictures.Finally,these images were preprocessed by graying,histogram equalization and normalization,so as to be sent to the neural network for higher recognition rate.(2)The method of recognizing children's smiling faces based on classicalconvolutional neural networks InceptionV3,ResNet50,Xception and MobileNet was studied.The four network models were fine-tuned by transfer learning,and the four network models were trained by adding FC layer and GAP layer.Through experiments,the convergence speed of each network model trained with FC layer is fast,so it can improve the classification performance of the original network and recognize the smiling face of children.At the same time,the smiley face recognition based on ResNet50 network model achieves the optimal performance in all aspects such as accuracy.However,the ResNet50 network model is large and has the most parameters,which means it has high requirements on the equipment.Therefore,on the premise of ensuring that the smiley face recognition rate will not have too much loss,combined with the structural characteristics of lightweight neural network,the ResNet50 compression was improved,and the ResNet50 network model with high recognition rate and less memory consumption and higher cost was finally realized.Compared with the original model,the recognition rate of the model is slightly improved,and the model is smaller,the parameters are reduced,the calculation cost is reduced,and the storage and calculation resources are saved.(3)Finally,a smile recognition system based on deep convolutional neural network for children is designed,which is composed of a face detection module,a face recognition module,a smiling face recognition module,a database module and a human-computer interaction interface module.After testing the function of each module,the smiling face recognition system designed in this thesis can achieve the effect of automatic face detection and recognition in real-time video and finally successfully extend the system to a primary school in Shanghai for the smiling face recognition and artificial intelligence technology experience.