摘要
运维人员正确的安全穿戴是确保电力作业安全的重要措施,计算机视觉深度学习算法为电力作业安全穿戴监管提供了一种新的手段。Faster R-CNN(faster region-based convolutional neural network)是一种有效的目标检测方法,在其基础上进行改进,首先以轻量级卷积神经网络EfficientNetV2(efficient network version 2)作为Faster RCNN的骨干,均衡提升算法的检测精度与速度;然后,在RPN(region proposal network)前引入注意力机制CBAM(convolutional block attention module)进一步提升检测精度。结合实际电力作业安全穿戴检测场景对该方法进行测试实验,结果表明,相对目前以VGG-16(visual geometry group-16)、ResNet-50(residual network-50)为骨干的Faster R-CNN算法而言,改进Faster R-CNN算法的检测精度和速度均得到提高,其平均精度均值(mean average precision,mAP)达到85.46%,检测速度达到47.82帧/s(frames per second,FPS),一定程度上提升了Faster R-CNN在电力作业安全穿戴检测中的应用能力。
The correct safety wearing of operation and maintenance personnel is an important measure to ensure the safety of electric power operations.The computer vision deep learning algorithm provides a new method for the supervision of safety wearing of electric power operations.Faster R-CNN(faster region-based convolutional neural network)is an effective object detection method,which is improved in this paper.Firstly,the lightweight convolutional neural network EfficientNetV2(efficient network version 2)was used as the backbone of Faster RCNN to improve the detection accuracy and speed of the algorithm in a balance manner.Then,CBAM(convolutional block attention module)was introduced before RPN(region proposal network)to further improve the detection accuracy.The method was tested in combination with the actual electric power operations safety wearing detection scenarios.The results showed that compared with current Faster R-CNN algorithm with VGG-16(visual geometry group-16)and ResNet-50(residual network-50)as the backbone,the detection accuracy and speed of the improved Faster R-CNN algorithm were improved,with the mean average precision(mAP)reaching 85.46%and the detection speed reaching 47.82 frames per second(FPS).The application capability of Faster R-CNN in electric power operation safety wearing detection is improved to a certain extent.
作者
李强
赵峰
李博强
许中平
余金沄
秦亮
LI Qiang;ZHAO Feng;LI Boqiang;XU Zhongping;YU Jinyun;QIN Liang(State Grid Information&Telecommunication Group Co.,Ltd.,Beijing 102211,China;School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China;Beijing State Grid Information&Telecommunication Group‐Accenture Information Technology Co.,Ltd.,Beijing 100052,China)
出处
《武汉大学学报(工学版)》
CAS
CSCD
北大核心
2024年第4期462-469,共8页
Engineering Journal of Wuhan University
基金
国网信息通信产业集团有限公司科技项目(编号:SGTYHT/19-JS-218)。