全局融合卷积神经网络的边缘分类的人脸性别识别  

Face gender identification based on edge classification of global fusion convolutional neural networks

作　　者：吴军[1] 邱阳 卢忠亮[1] WU Jun;QIU Yang;LU Zhongliang(School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西理工大学信息工程学院

出　　处：《现代电子技术》2019年第18期177-181,186,共6页Modern Electronics Technique

基　　金：国家自然科学基金青年项目(61701203);江西省教育厅科技项目(GJJ150642)~~

摘　　要：人脸性别识别是人脸识别领域研究的热门课题.为了进一步提高人脸性别识别的准确率,在传统的融合模型基础上,提出一种新型全局融合卷积神经网络模型(NFDCNN).在NFDCNN模型结构上,每两采样层之间的卷积层在卷积特征提取之前融合前一级的子采样特征,这种方法可以保留原始的特征信息同时与深层纹理融合,具有高度的还原度,缩小网络误差.NFDCNN模型分类函数在常规的Softmax上做了改进,引入了区域边缘分类函数AM-Softmax,该分类函数在归类上是以一块区域为界限来划分,挤压同类,扩大类间距离,缩小类内距离.实验是在不同的人脸数据集上采用该模型方法与其他先进方法对比,验证了提出的NFDCNN模型分类识别是有效的.A new global fusion convolutional neural network(NFDCNN) model based on the traditional fusion model is proposed to further improve the accuracy of face gender identification. On the structure of NFDCNN model,the convolutional layers between every couple of sampling layers are used to fuse the sub-sampling features in the front level before the convolution feature is extracted. This method can preserve the original feature information and fuse with the deep texture at the same time, which has high restoration degree to reduce network error. The NFDCNN model classification function has been improved on the conventional Softmax,and the regional edge classification function AM-Softmax is introduced. The classification function is divided by taking a region as a bound in classification,squeezing the same class,expanding the distance between classes and narrowing the distance within classes. The experiment is carried out by comparing the model method with other advanced methods on different face data sets. It is found in the verification that the classification and recognition of the proposed NFDCNN model are effective.

关 键 词：人脸性别识别 卷积神经网络 全局融合 纹理融合 边缘分类 模型验证 

分 类 号：TN915.34[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]