摘要
为了解决无人值守变电站由于点位众多所导致的难以对进站人员实时监控的问题,本文提出了一种针对进站人员实时监测的智能分类算法,首先采用级联Haar分类器实现对监控画面中人脸图像的捕获与分离,然后基于Face Net深度人脸识别模型完成对人脸图像的特征提取。在此基础上使用支持向量机算法完成对进站人员的智能分类:对于已知人员记录姓名以及进站时间,对于陌生人执行报警功能以及其他规定动作。在实际应用中的实验结果表明,调节算法超参数将获得不同的灵敏度与识别率,经过对超参数的微调,该算法的准确率达到90%左右。基于该算法开发的监控平台已部署到智能终端上,依靠边缘计算技术实现对无人值守变电站进站人员的自动识别,并在生产实践中取得了预期的效果。
In order to solve the problem that the unattended electric substation is difficult to monitor the incoming personnel in real time due to the large number of locations,this paper proposes an intelligent classification algorithm for real-time monitoring of incoming personnel. In the first step of our pipeline,cascade Haar classifier is adapted to capture and separate face images in monitoring screen. After that,features in face images are extracted based on Face Net deep face recognition model. On this basis,the support vector machine algorithm is used to classify incoming personnel intelligently: names and time are recorded when detecting known persons while alarm function and other actions could be executed when detecting strangers. The experimental results in practical applications show the fact that different sensitivity and precision could be obtained by tuning algorithm hyperparameters. The precision of this algorithm reaches about 90% after the process of fine-tuning. The monitoring platform which is developed based on this algorithm is now deployed on smart terminals,identifying entry personnel by edge computing technology automatically,which has achieved expected effect in practice.
作者
宗祥瑞
王洋
金尧
周斌
任新颜
庞玉志
ZONG Xiangrui;WANG Yang;JIN Yao;ZHOU Bin;REN Xinyan;PANG Yuzhi(State Grid Tianjin Electric Power company,Tianjin 300010,China)
出处
《电力大数据》
2020年第7期1-8,共8页
Power Systems and Big Data
关键词
无人值守变电站
支持向量机
人脸检测
人脸识别
智能监控
特征向量
unattended substation
support vector machine
face detection
face recognition
intelligent monitoring
feature vector
作者简介
宗祥瑞(1992),男,硕士,工程师,从事电力系统人工智能开发工作。
