摘要
当前输电线路巡检工作主要采用无人机巡检的方式来完成,利用人工智能技术助力巡检图像的缺陷识别已成为趋势。以往绝缘子缺陷检测研究大部分集中于绝缘子爆片类缺陷的检测,提出了一种改进的Cascade R-CNN算法,基于绝缘子缺陷数据建立绝缘子缺陷数据集。该算法主要针对绝缘子爆片、绝缘子电弧烧伤及绝缘子伞裙破损的多类型绝缘子缺陷联合检测。首先,针对目标比例小导致生成的anchor正负样本比例失衡的现象,将损失函数更改为Focal loss;然后引入了RoI Align方法以解决小目标的漏检;通过联合绝缘子串模型,将绝缘子串的检测区域设置为RoI,最后检测RoI区域中的绝缘子缺陷。实验结果显示,绝缘子爆片、绝缘子电弧烧伤及绝缘子伞裙破损三类缺陷检测mAP值可达到0.744,针对不同情形所提方法都能够有效地检测所设定的三类绝缘子缺陷目标,比Faster R-CNN方法更准确、更可靠。
Nowadays the inspection of transmission lines is mainly done by UVA inspection.It has become a trend to use artificial intelligence technology to assist the inspection of image defect identification.In the past,most researches on insulator defect detection focused on the detection of insulator missing faults.The author proposes an improved Cascade R-CNN algorithm to build insulator defect data sets based on insulator defect data.The algorithm is mainly aimed at the joint detection of multiple types of insulator defects,such as insulator missing faults,insulator was burned by electric shock and insulator with sheds damaged.First,the loss function is changed to Focal loss for the phenomenon that the proportion of the generated anchor positive and negative samples is unbalanced due to the relatively small number of targets,and then the RoI Align method is introduced to solve the missed detection of small targets.Finally,through the joint insulator string model,the detection area of the insulator string is set to RoI,and then it is used to detect the insulator defects in the RoI area.The experimental results show that the mAP value of the three types of defect detection for insulator missing faults,insulator burned by electric shock and insulator with sheds damaged can reach 0.744.The proposed method can effectively detect the three types of insulator defect targets set for different situations,which is more accurate and reliable than Faster R-CNN method.
作者
张欣
王红星
陈玉权
黄郑
沈杰
高小伟
ZHANG Xin;WANG Hongxing;CHEN Yuquan;HUANG Zheng;SHEN Jie;GAO Xiaowei(JiangSu Frontier Electric Technology,Nanjing 430075,China;Beijing Imperial Image Intelligent Technology,Beijing 430075,China)
出处
《电瓷避雷器》
CAS
北大核心
2022年第1期189-196,共8页
Insulators and Surge Arresters
基金
江苏方天电力技术有限公司科技项目“无人机智能巡检关键技术与三维平台应用”(编号:KJ201915)。
