摘要
面部表情是传递人类情感状态最直观的方式,通过分析面部表情,可以获得某人在某时刻的精神和身体状况。表情识别在人机通信、自动驾驶、医学等应用领域有着重要的应用价值,并且受到越来越多的关注。随着深度学习技术的发展,表情识别技术研究也从常规的图像处理方法转变为采用深度学习的方法,但是由于样本数量有限,以及硬件设备的限制,使得提高表情识别准确率的方法受到了一定的限制。文章主要对改进ALEXNET模型的表情识别算法进行研究,由于ALEXNET在人脸面部表情识别中准确率较低,因此在对ALEXNET网络进行深入研究的基础上,通过修改卷积核的大小以及卷积层的数量,增加注意力机制和残差块,以提高网络对人脸特征的提取能力,并将改进后的AlexNet模型应用于CK+及JAFFE数据集,进而在该数据集上取得了较好的识别准确率。
Facial expression is the most intuitive way to convey human emotional state.By analyzing facial expressions,one can obtain someone's mental and physical condition at a certain moment.Expression recognition has important application value in human-machine communication,autonomous driving,medicine and other application fields,and has received more and more attention.With the development of deep learning technology,the research of expression recognition technology has also changed from conventional image processing methods to deep learning methods.However,due to the limited number of samples and the limitation of hardware equipment,the method of improving the accuracy of expression recognition has been limited to a certain extent.This paper mainly studies the expression recognition algorithm of the improved AlexNet model.Since AlexNet has low accuracy in facial expression recognition,on the basis of in-depth research on the AlexNet network,by modifying the size of the convolution kernel and the number of convolutional layers,increase the attention mechanism and residual block to improve the network's ability to extract facial features,and apply the improved AlexNet model to the CK+,JAFFE dataset,and then achieve better recognition accuracy on this dataset.
作者
孙歌
王剑雄
欧琪
沈英杰
魏士磊
SUN Ge;WANG Jianxiong;OU Qi;SHEN Yingjie;WEI Shilei(Hebei University of Architecture,Zhangjiakou,Hebei 075024,China)
出处
《计算机应用文摘》
2023年第12期259-261,共3页
Chinese Journal of Computer Application
关键词
表情识别
AlexNet
深度学习
卷积层
facial expression recognition
AlexNet
deep learning
convolution layer