摘要
针对面部表情识别算法存在模型结构复杂、训练参数过多等问题,在对卷积神经网络各层分析的基础上,将迁移学习算法融入到面部表情识别模型中,从而对面部表情进行识别和分类.该文提出的算法首先通过数据增强的方法扩充面部表情图片的数量,然后将VGG16模型在ImageNet图像数据集上训练得到的权重参数,通过参数微调的方法,传递到面部表情识别模型中.最后采用全局平均池化层代替传统的全连接层,对表情图片通道信息进行求和运算实现降维,减少模型的网络参数.实验结果表明:提出的模型在KDEF数据集中取得了优异的识别效果,平均测试准确率达到96.23%.
In order to solve the problems of complex model structure and too many training parameters in facial expression recognition algorithm,based on the analysis of each layer of convolution neural network,the transfer learning algorithm is integrated into the facial expression recognition model to recognize and classify facial expressions.The algorithm proposed in this paper first expands the expression data images by means of data enhancement such as translation,rotation,scaling and filling,and then transfers the weight parameters trained by the VGG16 model on the ImageNet image data set to the facial expression recognition model by fine-tuning the parameters.Finally,the global average pooling layer is used to re⁃place the traditional full connection layer to reduce the dimension and reduce the network parameters of the model by summing the channel information of facial expressions.The experimental results show that the model proposed in this paper has achieved excellent recognition results in KDEF data sets,and the av⁃erage test accuracy is 96.23%.
作者
沈同平
黄方亮
许欢庆
SHEN Tong-ping;HUANG Fang-liang;XU Huan-qing(School of Medicine and Information Engineering,Anhui University of Chinese Medicine,Hefei 230012,China)
出处
《通化师范学院学报》
2023年第4期52-58,共7页
Journal of Tonghua Normal University
基金
安徽省高校优秀拔尖人才培育项目(gxyq2022026)
安徽省质量工程项目(2020jyxm1029,2021jyxm0801)
安徽高校自然科学研究重点项目(KJ2020A0443)
安徽中医药大学校级质量工程项目(2021zlgc046)
安徽中医药大学人文重点项目(2021rwzd20,2020rwzd07)
安徽中医药大学自然重点项目(2020zrzd18
2019zrzd11)。
关键词
表情识别
迁移学习
VGG16
facial expression recognition
transfer learning
VGG16
