摘要
为了解决传统卷积神经网络用于人脸表情识别准确率不高的问题,提出了一种基于改进深度AlexNet卷积神经网络的表情识别方法。该方法基于AlexNet网络的基本结构,采用单图形处理单元(Graphics Processing Unit,GPU)进行训练,减少了两层卷积层和一层全连接层,在每层卷积层后加上批标准化(Batch Normalization,BN)代替原来的局部归一化,并在全连接层后加上Dropout正则化进一步防止过拟合。与AlexNet模型相比,改进的网络结构更简单、复杂度低、参数量少,可以节省大量模型训练时间进行快速预测,且更不易过拟合,同时加快了模型收敛速度,提高了网络泛化能力。在Fer2013数据集以及CK+数据集上进行实验,结果表明,所提方法分别得到了68.85%和97.46%的识别率,较其他人脸表情识别方法的识别率有一定提高。
In order to solve the problem of low accuracy of traditional convolutional neural network(CNN)for facial expression recognition,an expression recognition method based on improved deep AlexNet CNN is proposed.Based on the basic structure of AlexNet network,this method uses single graphics processing unit(GPU)to train,thus reducing two layers of convolution layer and one layer of full connection layer,adding batch normalization(BN)after each layer of convolution layer to replace the original local normalization,and adding dropout regularization after the full connection layer to further prevent over fitting.Compared with AlexNet model,the improved network has simpler structure,lower complexity,and fewer parameter,and it can save a lot of model training time for fast predictions and is more difficult to overfit.Meanwhile,it can accelerate the convergence speed of the model and improve the generalization ability of the network.Experiments on Fer2013 data set and CK+data set show that the recognition rate of this method is 68.85%and 97.46%respectively,which is higher than other facial expression recognition methods.
作者
石翠萍
谭聪
左江
赵可新
SHI Cuiping;TAN Cong;ZUO Jiang;ZHAO Kexin(College of Communication and Electronic Engineering,Qiqihar University,Qiqihar 161006,China)
出处
《电讯技术》
北大核心
2020年第9期1005-1012,共8页
Telecommunication Engineering
基金
国家自然科学基金青年基金项目(41701479)
中国博士后科学基金项目(2017M621246)
黑龙江省科学基金项目(QC2018045)
黑龙江省博士后科学基金项目(LBH-Z17052)
黑龙江省省属高等学校基本科研业务费专项(135309342)
2019年省级大学生创新创业训练计划资助项目(201910232044)。
