轻量级深度学习电网作业安全检测算法研究  

Research on Lightweight Deep Learning Safety Detection Algorithm for Grid Operation

作　　者：刘加伟 陈新楚[1] 江灏[1,2] 缪希仁[1,2] 陈静[1,2] LIU Jiawei;CHEN Xinchu;JIANG HAO;MIAO Xinren;CHEN JING(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou,China,350108;Research Institute of Power System and Power Equipment,Fuzhou University,Fuzhou,China 350108)

机构地区：[1]福州大学电气工程与自动化学院,福州350108 [2]福州大学电力系统与装置产业研究院,福州350108

出　　处：《福建电脑》2023年第4期13-18,共6页Journal of Fujian Computer

基　　金：融合视觉感知与交互的架空输电线路无人机边云协同巡检关键技术研究(No.2022H6020)资助。

摘　　要：随着智能电网的迅速发展,在变电站现场安全作业检测时,时常发生作业人员安全帽、安全带、绝缘手套等佩戴异常情况。为了进一步保障电网安全,针对变电站现场环境复杂、人员较多、实时性要求高等情况,本文提出一种M-YOLOV5s算法,针对检测的实时性要求,将YOLOV5s的特征提取网络换成MobileNet V3 small网络,并引入自适应ACON激活函数。同时,针对现场小目标,在模型特征融合层中加入轻量级注意力机制CBAM,进一步提升模型性能。实验结果表明,采用本文所提出的方法进行电网现场作业安全检测,模型体积缩小至4.89MB,在极大提高模型检测速度的同时,模型精度仅损失3.12%,更有利于模型的前端部署。此外,该方法能够满足电网现场作业人员穿戴安全检测的实时性。With the rapid development of smart grids,abnormalities in the wearing of helmets,safety belts and insulated gloves by operators occur from time to time during the safety operation detection at substation sites.In order to further ensure the safety of the power grid,for the complex environment of substation sites,more personnel and high real-time requirements,this paper proposes an M-YOLOV5s algorithm,which replaces the feature extraction network of YOLOV5s with MobileNet V3 small network and introduces the adaptive ACON activation function for the real-time requirements of detection.Meanwhile,for small targets in the field,in the model feature fusion layer we add the lightweight attention mechanism CBAM to further improve the model performance.The experimental results show that the model size is reduced to 4.89 MB by using the method proposed in this paper for safety detection of grid field operations,and the model accuracy is only lost by 3.12%while greatly improving the model detection speed,which is more conducive to the front-end deployment of the model.In addition,the method can meet the real-time nature of the safety detection of grid field operation personnel wearing.

关 键 词：电网作业 穿戴安全检测 目标识别 轻量化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]