摘要
针对目前人脸表情识别(Facial Expression Recognition, FER)方法准确率低、模型大和识别时间长的问题,提出了一种基于卷积神经网络的通道注意力FER算法,在普通的卷积层中加入Xception网络中的可分离卷积网络,减少参数量和运算成本。在可分离卷积层的输出加入通道注意力Senet,实现对输出通道的权值按重要程度进行重新分配。引入Resnet网络中残差机制,减轻梯度消失现象。对设计的模型分别在CK+,RAF-DB数据集和FER2013数据集进行训练。实验结果显示,在CK+,RAF-DB数据集和FER2013数据集准确率分别提高至99.45%,78.10%和62.65%。模型参数量仅有1.8 MB,识别时间1.24 s。实现了更准、更快、更轻的FER。
To address the low accuracy, large model size and long recognition time of current facial expression recognition methods, a channel attention facial expression recognition algorithm based on convolutional neural network is proposed.A separable convolutional network in Xception network is added to the ordinary convolutional layer, which reduces the amount of parameters and the computational cost.The channel attention Senet is added to the output of the separable convolutional layer to realize the redistribution of the weight of the output channel according to the degree of importance.The residual error mechanism in Resnet network is introduced to reduce the phenomenon of gradient disappearance.The model designed is trained on CK+,RAF-DB dataset and FER2013 dataset respectively.The experimental results show that the accuracy of CK+,RAF-DB data set and FER2013 dataset is increased to 99.45%,78.10% and 62.65% respectively.The volume of model parameter is only 1.8 MB,and the recognition time is 1.24 s.A more accurate, faster and lightweight facial expression recognition is realized.
作者
张波
兰艳亭
李大威
牛兴龙
ZHANG Bo;LAN Yanting;LI Dawei;NIU Xinglong(School of Electrical and Control Engineering,North University of China,Taiyuan 030051,China)
出处
《无线电工程》
北大核心
2022年第1期148-153,共6页
Radio Engineering
基金
国家自然科学基金(61903343)
山西省科技攻关项目(20140311027-2)
中北大学17届研究生科技立项项目(20201772)。