基于红外图像目标检测与偏斜矫正的变压器套管发热缺陷检测方法  

Detection Method of Transformer Bushing Thermal Defect Based on Infrared Image Target Detection and Skew Correction

作　　者：张润 邱志斌 童志鹏 吴睿雯 唐智健 ZHANG Run;QIU Zhibin;TONG Zhipeng;WU Ruiwen;TANG Zhijian(Department of Energy and Electrical Engineering,Nanchang University,Nanchang 330031,China;Ji'an Power Supply Company,State Grid Jiangxi Electric Power Co.,Ltd.,Ji'an,Jiangxi 343000,China)

机构地区：[1]南昌大学能源与电气工程系,南昌330031 [2]国网江西省电力有限公司吉安供电分公司,江西吉安343000

出　　处：《南方电网技术》2024年第9期59-68,共10页Southern Power System Technology

基　　金：江西省“双千计划”创新领军人才长期(青年)项目(jxsq2019101071)。

摘　　要：变压器套管在红外巡检图像中占比较小,发热缺陷特征不明显,人工检测套管发热缺陷易受主观判断影响,且难以应对巡检产生的海量红外图像。为提高套管发热缺陷检测效率,提出了一种结合目标检测算法与图像偏斜矫正的变压器套管发热缺陷检测方法。首先,采用YOLOv7目标检测模型对套管目标进行识别与定位,引入SimAM注意力机制与高效解耦头对模型进行改进,提高套管目标的识别准确率与召回率。然后,对定位裁剪的套管目标进行图像偏斜矫正,提取中心区域温度特征信息进行发热缺陷诊断。实验结果表明:改进后模型对套管目标识别准确率为95.50%,召回率为97.14%,平均精度为98.30%,检测FPS为42帧/s,所提方法能精准定位套管目标并提取对应温度曲线,有效提高了套管发热缺陷检测效率。Transformer bushings account for a small proportion in infrared inspection images,and the characteristics of thermal defects are not obvious.Manual detection of thermal defects in bushings is easily affected by subjective judgments,and it is difficult to cope with the massive infrared images generated during inspections.In order to improve the detection efficiency of bushing thermal defects,a detection method for transformer bushing thermal defects combined with target detection algorithm and image skew correction is proposed.Firstly,the YOLOv7 object detection model is used to identify and locate bushing targets.The SimAM atten⁃tion mechanism and efficient decoupled head are introduced to optimize the model,improving the recognition accuracy and recall rate of bushing targets.Then,image skew correction is performed on the positioned and cropped bushing target,and temperature feature information in the central area is extracted for thermal defect diagnosis.The experimental results show that the improved model has an accuracy rate of 95.50%for bushing target recognition,a recall rate of 97.14%,an average accuracy of 98.30%,and a detection FPS of 42 frames per second.The proposed method can accurately locate bushing targets and extract corresponding temperature curves,effectively improving the efficiency of bushing thermal defect detection.

关 键 词：套管 红外图像 目标检测 温度曲线 发热缺陷 偏斜矫正 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM769[自动化与计算机技术—计算机科学与技术]