基于局部和全局特征的电力设备红外和可见光图像匹配方法  

Infrared and Visible Image Matching Method for Power Equipment Based on Local and Global Features

作　　者：冯旭刚[1,2,3] 阮善会 王正兵 安硕 张科琪 Feng Xugang;Ruan Shanhui;Wang Zhengbing;An Shuo;Zhang Keqi(Anhui Province Engineering Laboratory of Intelligent Demolition Equipment,Anhui University of Technology,Maanshan,243032,China;Anhui Engineering Research Center for Intelligent Applications and Security of Industrial Internet,Anhui University of Technology,Maanshan,243032,China;School of Electrical and Information Engineering,Anhui University of Technology,Maanshan,243032,China)

机构地区：[1]安徽省智能破拆装备工程实验室(安徽工业大学),马鞍山243032 [2]安徽省工业互联网智能应用与安全工程研究中心(安徽工业大学),马鞍山243032 [3]安徽工业大学电气与信息工程学院,马鞍山243032

出　　处：《电工技术学报》2025年第7期2236-2246,2305,共12页Transactions of China Electrotechnical Society

基　　金：安徽省高等学校自然科学研究重点项目(2023AH051107);安徽省智能破拆装备工程实验室开放基金项目(APELIDE2022A001);安徽省工业互联网智能应用与安全工程研究中心开放基金项目(IASII22-07);安徽高校自然科学研究重大项目(KJ2021ZD0042)资助。

摘　　要：针对电力设备红外和可见光图像匹配过程受图像局部灰度差异影响大,以及特征点描述和匹配困难的问题,该文提出了基于局部和全局特征的电力设备红外和可见光图像匹配方法。首先,利用多尺度角检测算法分别检测红外和可见光图像中的特征点,再使用不同尺度的曲率信息为每个特征点分配特征主方向(CAO);其次,分别构建每个特征点的部分灰度不变特征描述符(PIIFD)和全局上下文特征描述符;然后,将两种特征描述符的相似度进行加权融合,并利用双向匹配方法和随机抽样一致(RANSAC)方法筛选出正确的匹配点对;最后,得到图像间的仿射变换模型参数。实验结果表明:该文匹配方法与PIIFD、Log-Gabor直方图描述符(LGHD)和CAO匹配算法相比,正确匹配数显著增加,平均准确率较其他三种算法分别提高了50.71、27.62、11.11个百分点,平均重复度分别提高了27.69、28.81、19.18个百分点。The most intuitive manifestation of power equipment failure is temperature abnormality,through the fusion of infrared and visible images of power equipment can realize the analysis and detection of equipment operation status.Aiming at the problem that the matching process of infrared and visible images of power equipment is greatly affected by the local intensity difference of the images and the difficulty of feature point description and matching,an infrared and visible image matching method for power equipment based on local and global features is proposed.First,the infrared and visible images of power equipment undergo preprocessing through grayscale conversion and normalization.The Canny algorithm is then applied to extract prominent contour features.Secondly,a threshold is set to solve the problem of possible misestimation in k-cosine estimation of curvature.Then the feature points in the contour image are detected separately by the multi-scale corner detection algorithm based on arithmetic mean k-cosine curvature.The proposed curvature adaptive weighting-based principal direction assignment method is used to assign feature principal directions to each feature point to achieve scale and rotation invariance.Then the PIIFD and global context descriptor are constructed for each feature point,and the proposed similarity bidirectional matching method is utilized to complete the preliminary matching.Finally,the RANSAC algorithm is used to obtain the final matching results and thus the parameters of the affine transformation model between images.The experimental results show that the algorithm in this paper successfully matches 10 groups of images,and the average accuracy of the proposed method is 93.41%,and the average repetition rate is 33.56%,which verifies that the matching method in this paper can effectively match the infrared and visible images of power equipment.Compared with PIIFD,LGHD and CAO matching algorithms,the number of correct matching points is significantly increased,the average accuracy is impro

关 键 词：电力设备 图像匹配 红外和可见光图像 全局上下文描述符 特征相似度匹配 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]