引入感知模型的改进孪生卷积神经网络实现人脸识别算法研究  

Research on Inception Module Incorporated Siamese Convolutional Neural Networks to Realize Face Recognition

作　　者：徐先峰[1] 张丽 郎彬 夏振 XU Xian-feng;ZHANG Li;LANG Bin;XIA Zhen(College of Electronics and Control Engineering,Chang’an University,Xi’an,Shaanxi 710064,China)

机构地区：[1]长安大学电子与控制工程学院,陕西西安710064

出　　处：《电子学报》2020年第4期643-647,共5页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61201407,No.61473047);陕西省自然科学基础研究计划(No.2016JQ5130);陕西省重点研发计划(No.2019GY-002);长安大学中央高校基本科研业务费(No.300102328202);西安市智慧高速公路信息融合与控制重点实验室(No.ZD13CG46)。

摘　　要：针对非限定性条件下人脸识别困难问题,设计了一种引入感知模型的改进孪生卷积神经网络结构(Inception Module Incorporated Siamese Convolutional Neural Networks,IMISCNN),在充分利用孪生结构有效减少外界干扰并避免过拟合等优点的基础上,为其增加感知模型实现更丰富特征的提取.为寻找最优学习率引入了循环学习率策略,加速模型收敛.在CASIA-webface和Extended Yale B标准人脸数据库上的仿真实验表明,所提IMISCNN算法提升了人脸识别精度.To solve the problem of face recognition under unlimited conditions,a simple network structure named Inception Module Incorporated Siamese Convolutional Neural Networks(IMISCNN)was designed,which was suitable for small-scale data sets.On the basis of making full use of the Siamese structure to effectively reduce external interference and avoid over-fitting,inception module was incorporated to the Siamese network to extract richer features.Furthermore,a cyclical learning rate strategy was adopted to accelerate the convergence of the model.Simulation results on the CASIA-webface and Extended Yale B standard face database showed that the recognition accuracy of IMISCNN was significantly improved compared with other face recognition algorithms.

关 键 词：人脸识别 孪生卷积神经网络 感知模型 循环学习率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]