面向电力图像的混合特征配准算法

Hybrid Feature Alignment Method for Power System Imager

摘要: 电力设备长时间运行过程中，非正常的温升现象可能诱发电力设备失效，造成区域性供电中断，对工业生产、居民生活及社会经济运行带来重大损失。由于红外图像分辨率小，导致基于视觉的温度检测算法无法准确定位发热源位置，提出一种基于联合特征的多模态图像配准算法。首先，使用Canny算子对电力设备红外图像及可见光图像进行边缘轮廓提取增强其特征；然后，采用K-means聚类算法对红外图像进行目标分割，排除背景干扰，提升后续特征提取的有效性；最后，进行特征提取与特征匹配，基于匹配点对计算最优几何变换模型，实现电力设备异源图像配准。在Hpatches等数据集上的实验表明，本文算法相较于其他配准算法准确度和有效特征对均有较大提升，能结合测温算法对电力设备中异常温度源精准定位，辅助电力检修人员快速故障定位。

Abstract: During the long-term operation of power equipment, abnormal temperature rise may trigger equipment failure, causing regional power outages and significant losses to industrial production, residents’ lives, and socio-economic operations. Due to the low resolution of infrared images, vision based temperature detection algorithms cannot accurately locate the location of heat sources. A multi-modal image registration algorithm based on joint features is proposed. Firstly, the Canny operator is used to extract edge contours and enhance the features of infrared and visible light images of power equipment; Then, the K-means clustering algorithm is used to perform target segmentation on the infrared image, eliminate background interference, and improve the effectiveness of subsequent feature extraction; Finally, feature extraction and feature matching are performed, and the optimal geometric transformation model is calculated based on the matching point pairs to achieve heterogeneous image registration of power equipment. Experiments on datasets such as Hpatches have shown that our algorithm has significantly improved accuracy and effective feature pairs compared to other registration algorithms. It can be combined with temperature measurement algorithms to accurately locate abnormal temperature sources in power equipment and assist power maintenance personnel in quickly locating faults.