| 摘 要: 针对传统卷积神经网络(Convolutional Neural Network,CNN)在人脸表情识别中有效特征提取不够、泛化能力不强及识别准确性不高等缺点,研究选用具有更小卷积核和更深池化层的视觉几何组网络(Visual Geometry Group Network,VGGNet)进行人脸表情识别系统的设计。为了验证识别效果,在传统CNN和VGGNet框架下进行人脸表情识别系统的搭建,使用FER2013公开数据集进行训练和测试,并对测试结果进行对比分析。实验表明,传统CNN模型在人脸表情识别任务中的识别率仅为88%,而VGGNet则能够取得高达98%的识别率。此外,利用真人实际测试验证了研究搭建的VGGNet模型具有很好的表情识别效果。 |
| 关键词: 深度学习;卷积神经网络;VGGNet;人脸表情识别;FER2013 |
|
中图分类号: TP391
文献标识码: A
|
|
| Research on Face Expression Recognition Based on VGGNet |
|
LIAO Qingjiang, LIU Ting, ZHANG Xingyue, DONG Qi, LI Lele, LIU Jiahao
|
(School of In f ormation Engineering, Tianjin University of Commerce, Tianjin 300134, China)
3264358645@qq.com; liuting@tjcu.edu.cn; 3367328354@qq.com; 3347692550@qq.com; 2406697527@qq.com; 1912842862@qq.com
|
| Abstract: Given the shortcomings of traditional Convolutional Neural Networks ( CNN) in face expression recognition, such as insufficient effective feature extraction, weak generalization ability, and low recognition accuracy, this paper proposes to design a face expression recognition system by using Visual Geometry Group Network (VGGNet), which has smaller convolutional kernel and deeper pooling layer. To verify the recognition effect, the face expression recognition system is built under the framework of traditional CNN and VGGNet, the FER2013 public dataset is used for training and testing, and the testing results are analyzed and compared. The experiments show that the recognition rate of the traditional CNN model in the task of face expression recognition is only 88% , while the recognition rate of the VGGNet is as high as 98% . In addition, test with real persons verifies that the VGGNet model proposed in this paper has a very good expression recognition effect. |
| Keywords: deep learning; CNN; VGGNet; face expression recognition; FER2013 |
