基于多尺度特征融合Faster R-CNN的绝缘子自爆缺陷研究  

Research on Insulator Self-detonation Defects Based on Multi-scale Feature Fusion Faster R-CNN

作　　者：周自强 赵淳[1,2] 范鹏 ZHOU Zi-qiang;ZHAO Chun;FAN Peng(NARI Group Corporation(State Grid Electric Power Research Institute),Nanjing 211106,China;Wuhan NARI Limited Liability Company of State Grid Electric Power Research Institute,Wuhan 430074,China)

机构地区：[1]南瑞集团有限公司(国网电力科学研究院),江苏南京211106 [2]国网电力科学研究院武汉南瑞有限责任公司,湖北武汉430074

出　　处：《水电能源科学》2020年第11期187-189,44,共4页Water Resources and Power

基　　金：国家电网公司科技项目(524606180084)。

摘　　要：随着无人机巡检技术的发展,无人机拍摄的绝缘子图片数量呈指数增长,亟需一种高效的缺陷识别方法,为此提出一种改进后的深度学习方法。首先,采用多尺度特征融合方法改进传统的Faster R-CNN方法,实现绝缘子小目标的精准识别;然后,结合图像处理方法实现自爆绝缘子的识别和定位;最后,以某500kV输电线路采集的大量绝缘子图片作为数据集对方法进行验证。结果表明,所提方法适用于不同排列方式下的绝缘子缺陷检测,绝缘子自爆缺陷检测准确率为91.3%,检测效率较高。结果可为无人机巡检提供一定的技术支撑。With the development of UAV grid inspection technology,the number of insulator pictures is increasing exponentially,and it is in urgent need of an efficient defect detection method.Therefore,this paper proposes a self-exploding insulator defect localization method based on deep learning optimized.Firstly,the multi-scale merged strategy was used to improve the traditional Faster R-CNN method to achieve accurate identification of small targets of insulators.Then,the recognition and positioning of the self-detonating insulator were realized by image processing method.Finally,a large number of insulator pictures collected from a certain 500 kV transmission line were used as the data set to verify the method.The results show that the proposed method is suitable for defect detection of insulators in different arrangements,the detection accuracy of insulator self-detonation defect is 91.3%,and the detection efficiency is high.The research results can provide certain technical support for UAV grid inspection.

关 键 词：绝缘子 自爆缺陷 深度学习 图像处理 多尺度融合 

分 类 号：TM755[电气工程—电力系统及自动化]