多尺度智慧变电站全景巡检图像显著性区域检测模型  

Multi-scale Intelligent Substation Panoramic Inspection Image Saliency Region Detection Model

作　　者：张小东 王晶 吴相荣 史浩鹏 李瑞 ZHANG Xiaodong;WANG Jing;WU Xiangrong;SHI Haopeng;LI Rui(State Grid Gansu Electric Power Company,Lanzhou 730030,China;SGIT-UNI(Lanzhou)Cloud Date Technology Co.,Ltd.,Lanzhou,730050,China;Pingliang Power Supply Company of State Grid Gansu Electric Power Company,Pingliang 744000,China)

机构地区：[1]国网甘肃省电力公司,甘肃兰州730030 [2]国网思极飞天(兰州)云数科技有限公司,甘肃兰州730050 [3]国网甘肃省电力公司平凉供电公司,甘肃平凉744000

出　　处：《微型电脑应用》2025年第2期149-152,161,共5页Microcomputer Applications

摘　　要：为了准确发现变电站核心部件隐患并及时维修,防止重大事故出现,提出了多尺度智慧变电站全景巡检图像显著性区域检测模型。利用线性组合把初始数据整合为主分量,替换原始高维数据,获得模型内参变量的后验期望,去除噪声分量。通过目标识别不变矩的显著特征向量,选取核心函数改变空间,把特征矢量带入支持向量机内,得到合理极值。按照特征点匹配的效率与准确度,构建配准图像和参考图像显著区域位置关系,算出不超过像素点的正则化互相关数,剔除不匹配的特征点对,获得对应像素在图像内的坐标,实现显著性区域检测。通过实验证明所提方法能将巡检图像中背景与目标相互区分,且纹理细节清晰,检测准确性高。In order to find out the hidden dangers of the core components of the substation and repair them in time,and prevent major accidents,a multi-scale intelligent substation panoramic inspection image saliency region detection model is proposed.The linear combination is used to integrate the initial data into the main component,replace the original high-dimensional data,obtain the posterior expectation of the model’s internal parameters,and remove the noise component.Through the significant feature vector of the invariant moment of target recognition,the kernel function is selected to change the space,and the feature vector is brought into the support vector machine to obtain the reasonable extreme value.According to the efficiency and accuracy of feature point matching,the positional relationship between the saliency region of the registered image and the reference image is constructed,the regularized cross-correlation number that does not exceed the pixel points is calculated,the mismatched feature point pairs are eliminated,the coordinates of the corresponding pixels in the image are obtained,and the detection of the saliency region is realized.Experiments show that the proposed method can distinguish the background and the target in the patrol image,and the texture details are clear and the detection accuracy is high.

关 键 词：智慧变电站 全景巡检图像 显著性区域检测 图像预处理 区域背景分离 

分 类 号：TM63[电气工程—电力系统及自动化]