基于卷积神经网络的人脸表情识别  

Facial Expression Recognition Based On Convolutional Neural Network

作　　者：姚梦竹 黄官伟 YAO Meng-zhu;HUANG Guan-wei(CDHK,Tongji University,Shanghai 200092,China;School of Economics and Management,Tongji University,Shanghai 200092,China)

机构地区：[1]同济大学中德学院,上海200092 [2]同济大学经济与管理学院,上海200092

出　　处：《电脑知识与技术》2020年第16期19-23,共5页Computer Knowledge and Technology

摘　　要：表情识别在医学、商业和刑事侦查等领域中有着广泛的应用前景。针对表情识别技术的研究历时半个世纪,经历了由传统的手工提取特征向卷积神经网络自动提取特征的飞跃。卷积神经网络由于其自学习能力因而得到了广泛应用,但仍存在训练时间过长、参数量过大等问题。该文针对以上问题,在Xception神经网络的基础上简化了模型的网络层级,删除了传统神经网络的全连接层并使用深度可分离卷积替代传统的卷积层,构造了mini-Xception网络模型。通过在Fer2013公开人脸表情数据集上进行实验,取得了66%的识别精度。改进后的模型显著降低了训练参数量并缩短了训练时间,提高了模型的泛化能力。Facial expression recognition has a broad application prospect in the areas such as medicine,business and criminal inves⁃tigation.The research on facial expression recognition has been going on for half a century,the manual feature extraction has been improved to the automatic feature extraction based on the convolutional neural network(CNN).CNN has been widely used due to its self-learning characteristics,but there are still some problems such as too long training time and too many parameters.Aiming at the above problems,this paper simplifies the construction of the model Xception Neural Network,removes the full connection layer and uses the depthwise separable convolution to construct the model Mini-xception.Good results have been achieved on the DataSet Fer2013.The new model reduces the number of parameters and the training time,which contributes to the stronger general⁃ization ability of the model.

关 键 词：表情识别 卷积神经网络 批量归一化 图像分类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]