[1]张文海,陈春玲.基于改进 GhostNet 模型的表情识别研究[J].计算机技术与发展,2022,32(08):60-65.[doi:10. 3969 / j. issn. 1673-629X. 2022. 08. 010]
　ZHANG Wen-hai,CHEN Chun-ling.Research on Expression Recognition Based on Modified GhostNet Model[J].,2022,32(08):60-65.[doi:10. 3969 / j. issn. 1673-629X. 2022. 08. 010]
点击复制

基于改进 GhostNet 模型的表情识别研究()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
32
期数:
2022年08期
页码:
60-65
栏目:
图形与图像
出版日期:
2022-08-10

文章信息/Info

Title:
Research on Expression Recognition Based on Modified GhostNet Model
文章编号:
1673-629X(2022)08-0060-06
作者:
张文海; 陈春玲
南京邮电大学 计算机学院、软件学院、网络空间安全学院,江苏 南京 210023
Author(s):
ZHANG Wen-hai; CHEN Chun-ling
School of Computer Science,Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
关键词:
表情识别; 损失函数; 深度学习; 卷积神经网络; GhostNet
Keywords:
expression recognition; loss function; deep learning; convolutional neural network; GhostNet
分类号:
TP391. 4
DOI:
10. 3969 / j. issn. 1673-629X. 2022. 08. 010
摘要:
针对目前卷积神经网络应用到人脸表情识别任务上时,计算复杂、输入尺度过大、类间差异小和类内差异大等问题,提出了一种基于改进 GhostNet 模型的解决方法。 首先,结合 GhostNet 模型思想,设计了改进 GhostNet 模型来提取表情特征,比原网络模型拥有更好的轻量级特性,并解决了 Ghost 瓶颈层中可能导致信息丢失的问题。 其次,结合 Island 损失函数和 Circle 损失函数设计思想,设计并采用了基于余弦相似性的损失函数来指导神经网络的学习。 该方法可以在特征空间中减小类内差异,增大类间差异,从而提升特征判别能力。 在 FERplus 数据集上进行实验验证,基于改进 GhostNet 模型方法在模型参数量和计算量更少的前提下,依旧有着更高的识别准确率和更快的识别速度,效果优于 Softmax 交叉熵损失函数和基于余弦距离的损失函数等,更适合移动端和嵌入式设备的使用场景。
Abstract:
To solve the problems of the complex calculations,the large input scale,the small difference in between-class expressions andthe large difference in within-class expressions in facial expression recognition task,a solution based on modified GhostNet model is proposed. Firstly,a modified GhostNet model is designed to extract expression features based on the idea of GhostNet model,which hasbetter lightweight characteristics than the original network model,and the problem that the information may be lost in the Ghost bottlenecklayer is solved. Secondly, combined with the idea of Island loss function and Circle loss function, the loss function based on cosinesimilarity is designed and used to guide the learning of neural network. The method reduces the intra class difference and increases theinter class difference in the feature space,thereby improving the effect of feature discrimination. Experiments on FERplus data set show that the proposed method based on improved GhostNet model has higher recognition accuracy and faster recognition speed on the premiseof the fewer parameters and calculations. The effect is better than Softmax and the loss function based on cosine similarity. It is moresuitable for use scenarios of mobile terminals and embedded devices.

相似文献/References:

[1]刘洋 季桂树 王刚.表情识别中基于轴对称的人眼定位方法[J].计算机技术与发展,2008,(12):47.
　LIU Yang,JI Gui-shu,WANG Gang.Eyes Location Method Based on Axial Symmetry Using for Expressions Recognition[J].,2008,(08):47.
[2]陈开云 高珏 孙柏林[] 许华虎[].基于混合特征提取的人脸情感识别研究[J].计算机技术与发展,2012,(02):96.
　CHEN Kai-yun,GAO Jue,SUN Bai-lin,et al.Studies on Facial Expression Recognition Based on Hybrid Features Extraction[J].,2012,(08):96.
[3]徐龙飞,郁进明.不同优化器在高斯噪声下对 LR 性能影响的研究[J].计算机技术与发展,2020,30(03):7.[doi:10. 3969 / j. issn. 1673-629X. 2020. 03. 002]
　XU Long-fei,YU Jin-ming.Study on Influence of Different Optimizers on Performance of LR under Gaussian Noise[J].,2020,30(08):7.[doi:10. 3969 / j. issn. 1673-629X. 2020. 03. 002]
[4]罗正军,柯铭菘,周德群.基于改进型 LSTM 的文本情感分析模型研究[J].计算机技术与发展,2020,30(12):40.[doi:10. 3969 / j. issn. 1673-629X. 2020. 12. 007]
　LUO Zheng-jun,KE Ming-song,ZHOU De-qun.Research on Text Sentiment Analysis Model Based on Improved LSTM[J].,2020,30(08):40.[doi:10. 3969 / j. issn. 1673-629X. 2020. 12. 007]
[5]张素智,吴玉红,常 俊.基于改进 AlexNet 卷积神经网络的轮胎图像识别[J].计算机技术与发展,2021,31(07):182.[doi:10. 3969 / j. issn. 1673-629X. 2021. 07. 030]
　ZHANG Su-zhi,WU Yu-hong,CHANG Jun.Tire Damage Image Recognition Based on Improved AlexNetConvolutional Neural Network[J].,2021,31(08):182.[doi:10. 3969 / j. issn. 1673-629X. 2021. 07. 030]
[6]周 婕,马明栋.基于改进的 ResNet 网络的人脸表情识别[J].计算机技术与发展,2022,32(01):25.[doi:10. 3969 / j. issn. 1673-629X. 2022. 01. 005]
　ZHOU Jie,MA Ming-dong.Facial Expression Recognition System Based on Improved ResNet[J].,2022,32(08):25.[doi:10. 3969 / j. issn. 1673-629X. 2022. 01. 005]
[7]姜丽莉,黄承宁.融合注意力机制改进残差网络的表情识别方法[J].计算机技术与发展,2022,32(05):42.[doi:10. 3969 / j. issn. 1673-629X. 2022. 05. 007]
　JIANG Li-li,HUANG Cheng-ning.An Expression Recognition Method Based on Fusion of Attention Mechanism and Improved Residual Network[J].,2022,32(08):42.[doi:10. 3969 / j. issn. 1673-629X. 2022. 05. 007]
[8]方承志,倪梦媛,唐 亮.基于残差网络及笔画宽度变换的场景文本检测[J].计算机技术与发展,2023,33(01):49.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 008]
　FANG Cheng-zhi,NI Meng-yuan,TANG Liang.Scene Text Detection Based on Residual Network andStroke Width Transform[J].,2023,33(08):49.[doi:10. 3969 / j. issn. 1673-629X. 2023. 01. 008]
[9]张 浩,陈亚南,杨 璐,等.基于注意力机制与残差网络的掌纹脊线距离估计[J].计算机技术与发展,2023,33(07):61.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 009]
　ZHANG Hao,CHEN Ya-nan,YANG Lu,et al.Palmprint Ridge Distance Estimation Based on Attention Mechanism and Residual Network[J].,2023,33(08):61.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 009]
[10]王 彪,毋 涛.基于卷积神经网络的面料检索系统[J].计算机技术与发展,2023,33(09):52.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 008]
　WANG Biao,WU Tao.Fabric Retrieval System Based on Convolutional Neural Network[J].,2023,33(08):52.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 008]

更新日期/Last Update: 2022-08-10