基于改进的Faster R-CNN网络的绝缘子识别算法  

Insulator fault identification algorithm based on improved Faster R-CNN network

作　　者：徐武[1] 高寒 王欣达 张强 秦浩然 吴庆畅[3] XU Wu;GAO Han;WANG Xin-da;ZHANG Qiang;QIN Hao-ran;WU Qing-chang(College of Communication and Electronic Engineering,Yunnan Minzu University,Kunming 650031,China;China Petroleum Transportation Corporation,Urumqi 830014,China;Lanmei Vocational and Technical College,Yunnan Minzu University,Kunming 650500,China)

机构地区：[1]云南民族大学电气信息工程学院,云南昆明650500 [2]中国石油运输公司,乌鲁木齐830014 [3]云南民族大学澜湄职业技术学院,云南昆明650500

出　　处：《云南民族大学学报（自然科学版）》2023年第5期595-600,共6页Journal of Yunnan Minzu University:Natural Sciences Edition

基　　金：国家自然科学基金(U1802271);2021年云南省专业学位研究生案例库建设项目;校级教学质量与教学改革工程项目(2020JGYB-04)。

摘　　要：在绝缘子检测过程中,航拍绝缘子图像中存在多个绝缘子串,且目标较小,背景复杂,导致绝缘子特征极其不明显.为解决以上问题,从提高网络模型对于小目标和多目标物体检测能力的角度对Faster R-CNN进行改进.首先,选用对提取特征信息利用率更高的GoolgeNet进行特征提取,利用Inception结构高效地利用计算资源;其次,针对一张航拍图像中有多个绝缘子目标,且大小不一,特征不易提取,对Inception结构进行改进;最后,在通用的数据集上进行测试,实现对绝缘子的精确检测.实验结果表明:该方法与传统的Faster R-CNN算法相比,准确率提高了约6%,召回率提升了约8%,能够较好地识别出绝缘子,且对于多目标和小目标的检测效果良好.In view of the small single object in the insulator string,the overall characteristics of the insulator are not obvious when the background is complex,and sometimes the whole insulator in aerial photography is small,the Fas⁃ter R-CNN is improved from the perspective of improving the detection ability of the network model for small and multi-target objects.Firstly,goolgenet,which has a higher utilization of extracted feature information,is selected for feature extraction,and the concept structure is used to make efficient use of computing resources;Secondly,aiming at the fact that there are multiple insulator targets in an aerial image,which are different in size and difficult to ex⁃tract features,the concept structure is improved;Finally,the test is carried out on the universal data set of the net⁃work to realize the accurate detection of insulators.The experimental results show that compared with the traditional Faster R-CNN algorithm,the accuracy of this method is improved by about 6%,and the recall rate is improved by about 8%,which can better identify insulators,and the detection effect for multi-target and small target is good.

关 键 词：绝缘子 区域建议网络 INCEPTION GoogleNet 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]