基于图像识别的输电线路轨道运输装备安全检测系统  

Safety detection system of rail transportation equipment for transmission lines based on image recognition

作　　者：王海燕 侯康 WANG Haiyan;HOU Kang(Huizhou Power Supply Bureau,Guangdong Power Grid Co.,Ltd.,Huizhou,Guangdong 516003,China;School of Mathematical Sciences,Soochow University,Suzhou,Jiangsu 215031,China;Kunshan Industrial Technology Research Institute Co.,Ltd.,Suzhou,Jiangsu 215316,China;Institute of Semiconductors,Chinese Academy of Sciences,Beijing 100083,China)

机构地区：[1]广东电网有限责任公司惠州供电局,广东惠州516003 [2]苏州大学数学科学学院,江苏苏州215031 [3]昆山市工业技术研究院有限责任公司,江苏苏州215316 [4]中国科学院半导体研究所,北京100083

出　　处：《江苏大学学报（自然科学版）》2024年第3期323-329,共7页Journal of Jiangsu University：Natural Science Edition

基　　金：中国南方电网有限责任公司科技项目(031300KK52190154);海南省重点研发计划项目(ZDYF2021GXJS213)。

摘　　要：为了提高林区山地输电线路轨道运输装备运行安全性,搭建基于图像识别的轨道运输装备安全检测系统.首先给出整个林区山地输电线路轨道运输装备电控系统;其次介绍了感知模块系统所用到的各类传感器;然后基于拆分注意力网络和自校准卷积的融合,采用Faster-RCNN算法得到更好的特征提取,并采用此改进的Faster-RCNN算法进行装备周围人员识别试验;最后基于QT开发输电线路轨道运输装备远程控制软件,并实现对装备的远程操控.结果表明:改进的Faster-RCNN算法在林区山地强光照和复杂环境下能够大幅度提高识别装备周围人员的准确性,图像识别平均精度均值mAP可达87.13%,高于常规Faster-RCNN的74.35%和级联Faster-RCNN的76.28%,充分证明改进的Faster-RCNN算法具备优良识别能力,保障林区山地输电线路轨道运输装备安全运行.To improve the operational safety of rail transportation equipment for mountainous transmission lines in forest areas,the safety detection system of rail transportation equipment based on image recognition was established.The electronic control system of rail transportation equipment for mountainous transmission lines in the forest areas was provided,and the various sensors used in the perception module system were introduced.The Faster-RCNN algorithm based on the combination of split-attention networks and self-calibration convolutions was applied to obtain better feature extraction,and the improved Faster-RCNN algorithm was used for the surrounding personnel recognition experiments.The remote control software of rail transportation equipment for transmission lines was developed based on QT,and the remote control of the equipment was realized.The results show that the improved Faster-RCNN algorithm can significantly improve the accuracy of identifying personnel around equipment in strong lighting and complex environments of forest areas.The mean average precision of image recognition can reach 87.13%,which is 74.35%higher than conventional Faster-RCNN and 76.28%higher than cascaded Faster-RCNN.The results fully prove that the improved Faster-RCNN algorithm has excellent recognition ability and ensures the safe operation of railway transportation equipment for mountainous transmission lines in forest areas.

关 键 词：输电线路建设 轨道运输装备 图像识别 感知系统 安全检测 改进的Faster-RCNN 拆分注意力网络 自校准卷积 

分 类 号：TP23[自动化与计算机技术—检测技术与自动化装置] TP391.4[自动化与计算机技术—控制科学与工程]