[1]张栋昱,赵 磊.融合注意力机制改进 ResNet 的人脸表情识别[J].计算机技术与发展,2023,33(05):130-137.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 020]
　ZHANG Dong-yu,ZHAO Lei.Improved Facial Expression Recognition in ResNet by Integrating Attention Mechanism[J].,2023,33(05):130-137.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 020]

点击复制

融合注意力机制改进 ResNet 的人脸表情识别()

分享到：

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

卷:

33

期数:

2023年05期

页码:

130-137

栏目:

人工智能

出版日期:

2023-05-10

文章信息/Info

Title:

Improved Facial Expression Recognition in ResNet by Integrating Attention Mechanism

文章编号:

1673-629X(2023)05-0130-08

作者:

张栋昱; 赵 磊

武汉大学 国家网络安全学院,湖北 武汉 430072

Author(s):

ZHANG Dong-yu; ZHAO Lei

School of National Cyber Security,Wuhan University,Wuhan 430072,China

关键词:

人脸表情识别; 深度学习; 残差网络; 卷积注意力机制; Dropout

Keywords:

facial expression recognition; deep learning; residual network; convolutional attention mechanism; Dropout

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 05. 020

摘要:

鉴于现有人脸表情识别方法在表情识别过程中存在的诸多痛点,比如对有效特征提取不够、泛化能力不强、识别准确性不高等,提出了一种改进残差网络的人脸表情识别方法。 首先,引入卷积注

意力机制,对网络中间的特征图进行重构,强调重要特征,抑制一般特征;其次,使用激活函数 PReLU 替换 ResNet 中原有的 ReLU,在提高模型拟合复杂数据能力的同时,避免出现在负值区域的梯度永远为 0,进而导致模型训练时无法执行反向传播的问题;然后,在网络输出层的avgpool 与 fc 之间加入 Dropout 抑制过拟合,以进一步增加网络模型的鲁棒性与泛化性;最后,在公开数据集 CK+上的仿真实验结果表明,该方法的准确识别率达到 96. 12% 。 与现有多种经典算法,以及 baseline 算法即 ResNet101 相比,改进的网络模型具有更好的识别效果,证明了该方法的有效性与优异性。

Abstract:

In view of many pain points in the process of expression recognition of the existing facial expression recognition methods,suchas the extraction of effective features is not enough, generalization ability is not strong, recognition accuracy is not high, a facialexpression recognition method with improved residual network is proposed. Firstly,the convolutional attention mechanism is introducedto reconstruct the feature map in the middle of the network,emphasizing important features and suppressing general features. Secondly,the activation function PReLU is used to replace the original ReLU in ResNet,which can not only improve the ability of the model to fitcomplex data, but also avoid the problem that the gradient in the negative?

area is always zero, which leads to the failure ofbackpropagation in model training. Then, Dropout is added between avgpool and fc in the output layer of the network to suppressoverfitting,

so as to further increase the robustness and generalization of the network model. Finally,the simulation experiments on theopen dataset CK+ show that the accurate recognition rate of the proposed method reaches 96. 12% . The improved network model hasbetter recognition effect than the existing classical algorithms and baseline algorithm,namely ResNet101,which proves the effectivenessand excellence of the proposed method.

常用功能