电厂安全帽佩戴安全性监控的智能图像处理方法  

Intelligent Image Processing Technology for Safety Helmet Wearing in Power Plant

作　　者：薛明华 艾春美 律慧瑾 苏明旭[2] XUE Minghua;AI Chunmei;LYU Huijin;SU Mingxu(Shanghai Minghua Electric Power Science Co.,Ltd.,Yangpu District,Shanghai 200090,China;University of Shanghai for Science and Technology,Yangpu District,Shanghai 200093,China)

机构地区：[1]上海明华电力科技有限公司,上海市杨浦区200090 [2]上海理工大学,上海市杨浦区200093

出　　处：《中国电机工程学报》2022年第9期3346-3353,共8页Proceedings of the CSEE

基　　金：国家电力投资集团科技研究项目(SH201812)。

摘　　要：立足于电厂安全管理现状,针对部分人员不按规定佩戴安全帽进入电厂作业区域造成安全隐患问题,对基于图像智能处理的安全帽佩戴检测技术开展研究。设计一种安全帽佩戴检测方法,通过掩膜区域卷积神经网络(mask region convolution neural network,Mask R-CNN)深度学习算法对作业人员图像分析,采集5000张安全帽佩戴照片样本作为训练图集,对其进行预处理,再由改进的特征金字塔网络算法(feature pyramid networks,FPN)进行神经网络训练。对于测试集500张图像分析结果表明,结合改进的FPN算法和ResNet-101作为主干网络的Mask R-CNN模型能够有效实现对安全帽佩戴与否及佩戴错误的检测,模型精确率为0.971,召回率为0.973,均值平均精确度为0.970,获得较准确的电厂应用场景下安全帽佩戴安全性检测效果。Under the current situation of safety management in power plant,a safety helmet wearing detection technology based on intelligent processing of images was investigated in order to reduce the potential safety risk caused by some personnel entering the operation area of the power plant without wearing safety helmets as required.A safety helmet wearing detection method was designed,where the mask region convolution neural network(mask R-CNN)deep learning algorithm was employed to analyze the image of operators.Five thousand safety helmet wearing photos were collected as the training set.After preprocessing,it experienced the neural network training using the improved feature pyramid network algorithm(FPN).For the test data of 500 images,the analysis results show that the Mask R-CNN model combined with the improved FPN algorithm and ResNet-101 as the backbone network can effectively distinguish whether the helmet is worn or not,as well as the inappropriate wearing cases.Consequently,the improved model can yield an accuracy rate of 0.971,a recall rate of 0.973,and an average accuracy of 0.970,which permits a better safety detection effect of helmet wearing in power plants.

关 键 词：电厂 安全性 智能监控 图像处理 卷积神经网络 

分 类 号：TM62[电气工程—电力系统及自动化]