基于红外与可见光图像融合的GIS设备气体泄漏识别研究  

GIS gas leakage recognition based on infrared and visible image fusion

作　　者：袁建华[1] 陈广生 张天宇 黄淘 陈轩 Yuan Jianhua;Chen Guangsheng;Zhang Tianyu;Huang Tao;Chen Xuan(College of Electrical Engineering and New Energy,Three Gorges University,Yichang 443002,China;Jingzhou Power Supply Company,State Grid Hubei Electric Power Co.,Ltd.,Jingzhou 434020,China)

机构地区：[1]三峡大学电气与新能源学院,宜昌443002 [2]国网湖北省电力有限公司荆州供电公司,荆州434020

出　　处：《国外电子测量技术》2024年第12期231-239,共9页Foreign Electronic Measurement Technology

基　　金：煤燃烧国家重点实验室开放基金(FSKLCCA1607);国网湖北省电力有限公司科技项目(B715J023001A)资助。

摘　　要：针对单一传感器识别GIS设备气体泄漏故障成像辨识度低的缺陷,提出了一种基于高斯滤波-VGG19的图像融合方法。首先,在红外和可见光图像的预处理阶段进行图像配准,然后利用高斯滤波法将配准的图像分解为低秩基础层和显著细节层;其次,对低秩基础层图像采用加权平均法进行处理,形成融合基础层,对显著细节层图像,引入VGG19网络提取多层特征子图像后,聚集多层融合策略获取显著细节层的融合结果,形成融合细节层;最后,通过重构求和策略将两个融合后的子图像结合起来得到最终完整融合图像。将融合后的图像输入YOLOv5气体泄漏检测模型中,进而实现气体泄漏状态识别。通过仿真实验验证了所提方法在GIS设备气体泄漏识别方面的有效性。Aiming at the defect of low recognition degree of single sensor to recognize gas leakage fault imaging of GIS equipment,an image fusion method based on Gaussian filter-VGG19 is proposed.Firstly,image alignment is performed in the preprocessing stage of infrared and visible images,and then the aligned images are decomposed into low-rank base layer and significant detail layer using Gaussian filtering.Secondly,the low-rank base layer images are processed using weighted mean method to form the fusion base layer,and for the significant detail layer images,a VGG19 network is introduced to extract the multi-layer feature sub-images,and then the multi-layer fusion strategy is aggregated to obtain the fusion results of the salient detail layer to form the fusion detail layer.Finally,the two fused sub-images are combined by the reconstruction and summation strategy to obtain the final complete fused image.The fused image is input into the YOLOv5 gas leakage detection model to realize gas leakage state recognition.Simulation experiments verify the effectiveness of the proposed method in recognizing gas leakage in GIS equipment.

关 键 词：高斯滤波法 气体泄漏识别 VGG19 图像融合 YOLOv5 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM595[自动化与计算机技术—计算机科学与技术] TN219[电气工程—电器]