摘要
人脸识别技术是深度学习的一个重要领域。为了克服深度神经网络的复杂结构和训练以及softmax层万局逼近能力的缺陷,文章基于ShuffleNet和集成随机权向量函数连接网络分类器(Random Vector Functional-Linknet,RVFL),探索了一种基于轻量级深层神经网络的人脸识别模型,在改进识别精度的同时,降低深度神经网络训练的复杂度,并面向具有不同安全等级要求的基建、营销、运检、安监、信息通信等电网核心业务领域,基于所构建的模型开发了通用的人员安全管控平台,构建多渠道的实名鉴别服务、多级别的身份管理服务、多元化的身份凭证管理服务以及统一身份认证服务整体解决方案,解决了人员真实身份验证和人员安全行为管控问题。实验表明平台能够简化工作流程,提高运行效率,降低管理成本。
Face recognition technology is an important area of deep learning. In order to overcome the complex structure and training of deep neural networks and the shortcomings of universal approaching ability of softmax layer, based on ShuffleNet and ensemble random vector function-link net (RVFL), this paper explores a face recognition model based on lightweight deep neural network, which improves the recognition accuracy and reduces the complexity of neural network training. For the grid core business areas, such as infrastructure, marketing, transportation inspection, security supervision, information communication, etc., based on the built model, a universal personnel safety management platform was developed, which built a multi-channel real-name authentication service, a multi-level identity management service, a diversified identity certificate management service and an unified identity authentication service to solve the personnel identity verification and personnel safety behavior control. Experiments show that the platform can simplify work flow, improve operational efficiency, and reduce management costs.
作者
杨德胜
马冬
YANG Desheng;MA Dong(State Grid Communication Industry Group Co., Ltd., Anhui Jiyuan Software Co., Ltd., Hefei 230088, China)
出处
《电力信息与通信技术》
2019年第6期1-7,共7页
Electric Power Information and Communication Technology
基金
国家电网有限公司总部科技项目资助“电网核心业务可信身份认证关键技术研究”(52110418001L)
作者简介
杨德胜(1982-),男,高级工程师,从事电力大数据、人工智能、移动互联和网络空间安全方面的研究工作,459099034@qq.com;马冬(1984-),男,工程师,从事电力大数据、人工智能、移动互联和网络空间安全方面的研究工作。
