基于改进YOLOv7的变电站套管红外图像智能诊断模型构建及应用  

Construction and application of intelligent diagnosis model of infrared image of substation casing based on improved YOLOv7

作　　者：钟志明[1,2] ZHONG Zhiming(Dongguan Power Supply Bureau of Guangdong Power Grid Company,Dongguan 523120,China;School of Electric Power,South China University of Technology,Heyuan 517400,China)

机构地区：[1]广东电网公司东莞供电局,东莞523120 [2]华南理工大学电力学院,河源517400

出　　处：《国外电子测量技术》2025年第1期140-147,共8页Foreign Electronic Measurement Technology

摘　　要：为提升变电站套管红外图像智能诊断效果,提出构建一个基于改进YOLOv7网络的变电站套管红外图像智能诊断模型。首先,在YOLOv7基础检测网络的基础上对其特征提取模块进行改进,以提升其目标检测速率;然后在YOLOv7网络中引入卷积注意力机制模块注意力机制,以提高网络检测精度;最后将改进的YOLOv7网络用于变压器套管红外图像的智能诊断。结果表明:在参数量为1200时,改进YOLOv7模型的平均精度值、每秒帧数和每秒浮点运算次数分别为96.6%、30帧/s和8次,均优于其他检测模型。将本模型部署至智能诊断系统中后,可有效提升变电站套管红外图像目标检测速率和精度,证明本模型可实现套管红外图像发热缺陷故障诊断,具备有效性。In order to improve the intelligent diagnosis effect of substation casing infrared image,an intelligent diagnosis model of substation casing infrared image based on improved YOLOv7 network is proposed.Firstly,the feature extraction module is improved based on the YOLOv7 basic detection network to improve its target detection rate;then the Convolutional Block Attention Module(CBAM)attention mechanism is introduced into the YOLOv7 network to improve the network detection accuracy;and finally,the improved YOLOv7 network is used for intelligent diagnosis of transformer casing infrared image.The results show that the average precision(AP)value,frames per second(FPS)and floating point operation per second were 96.6%,30 frames/s and 8 times for the number of parameters of 1200,respectively,which are better than the other detection models.After the model is deployed in the intelligent diagnosis system,it can effectively improve the detection rate and accuracy of the infrared image target of the substation casing,and prove that the model can realize the fault diagnosis of the casing infrared image heating defect,and be effective.

关 键 词：YOLOv7网络 变压器套管 红外图像 目标检测 智能诊断 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]