一种基于改进稠密卷积神经网络的表情识别方法  

Expression recognition method based on improved dense convolutional neural network

作　　者：戴沁璇 罗晓曙[1] 蒙志明 黄苑琴 DAI Qinxuan;LUO Xiaoshu;MENG Zhiming;HUANG Yuanqin(College of Electronic Engineering,Guangxi Normal University,Guilin 541000,China;Innovation and Entrepreneurship Institute,Guangxi Normal University,Guilin 541000,China)

机构地区：[1]广西师范大学电子工程学院,广西桂林541000 [2]广西师范大学创新创业学院,广西桂林541000

出　　处：《现代电子技术》2022年第9期29-34,共6页Modern Electronics Technique

基　　金：广西科技重大专项(桂科AA18118004);广西人文社会科学发展研究中心科学研究工程·创新创业专项(重大委托项目)(ZDCXCY01)。

摘　　要：人的表情包含大量信息,可用于显示人的很多情感状态,例如疲劳和疼痛的表情等。卷积神经网络(CNN)是一种识别人脸表情的有效方法,它可以同时执行特征提取和分类,并可以自动发现数据中的多个表情特点。针对卷积神经网络参数大以及传统表情识别方法准确率不高的问题,提出一种基于改进的稠密卷积神经网络的面部表情识别模型。首先通过使用Gabor滤波器初始化第一层卷积层;然后采用一种对数线性函数(LLU)进行网络优化,该模型中的特征重用和参数压缩技术提高了网络的学习能力,大大减少了模型参数;最后基于此模型设计了一个表情识别系统,该系统能够准确地识别照片上的表情和在线识别人脸表情。实验结果表明,该模型可以显著提高三个表情数据集的准确率,并能很好地识别人脸表情。Human expression contains a lot of information,which can be used to display many emotional states,for example,fatigue,pain,etc.The convolutional neural network(CNN)is an effective method for facial expression recognition(FER).It can perform feature extraction and classification at the same time,and can automatically discover multiple expression features in the data.In view of the large parameters of CNN and low accuracy of the traditional expression recognition methods,an FER model based on improved dense CNN is proposed.The Gabor filter is used to initialize the first convolutional layer first,and then a logarithmic linear unit(LLU)is used to optimize the network.The feature reuse and parameter compression technology of the model improve the learning ability of the network and greatly reduce the model parameters.An expression recognition system based on this model is designed,which can accurately recognize the expression on the photo and recognize the facial expression online.The experimental results show that the model can significantly improve the accuracy of the three expression data sets,and can recognize facial expressions well.

关 键 词：人脸表情识别 改进稠密卷积神经网络 卷积层初始化 GABOR滤波器 激活函数 表情识别系统 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]