改进型的MobileNet的轻量级人脸表情识别方法  

Improved mobileNet for lightweight facial expression identification

作　　者：吴振荣 邱卫根[1] 张立臣[1] WU Zhen-rong;QIU Wei-gen;ZHANG Li-chen(School of Computers,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学计算机学院,广州510006

出　　处：《信息技术》2024年第8期44-50,58,共8页Information Technology

基　　金：国家自然科学基金项目(61873068)。

摘　　要：为了解决目前轻量级卷积神经网络MobileNet应用于人脸表情识别准确率不高、实时性差和时空负载大等问题,文中提出了一种改进型的MobileNet的轻量级人脸表情识别方法。该方法在MobileNet X的基础上,引入SE注意力模块并针对表情图像的特点对深度卷积层和网络结构进行优化,避免了信息丢失和神经元“坏死”问题,提高了模型的人脸表情识别率。与MobileNet X模型相比,改进后的网络模型复杂度低、识别精度高。在Fer2013人脸表情数据集上的实验证明,文中方法得到了73.54%的识别率,较其他表情识别方法在识别率和时间效率上都有一定提高。In order to solve the problems of low accuracy,poor real⁃time performance and large space⁃time load in the application of lightweight convolutional neural network MobileNet to facial expression recogni⁃tion,this paper proposes an improved MobileNet lightweight facial expression recognition method.Based on MobileNet X,this method introduces the SE attention module and optimizes the deep convolution layer and network structure according to the characteristics of the expression image,avoiding the problem of informa⁃tion loss and neuron“necrosis”,and improving the recognition rate of the model’s facial expression.Com⁃pared with MobileNet X model,the improved network model has low complexity and high recognition accu⁃racy.The experiment on Fer2013 facial expression dataset shows that the recognition rate of this method is 73.54%,which is higher than other facial expression recognition methods in recognition rate and time effi⁃ciency.

关 键 词：人脸表情识别 卷积神经网络 轻量级 注意力机制 深度可分离卷积 

分 类 号：TP319.4[自动化与计算机技术—计算机软件与理论]