针对传统基于聚类的红外图像异常检测方法对电力设备红外图像多层分割效果较差, 异常检测有效性较低等问题, 提出了一种核猫群电力红外图像异常检测方法, 通过核猫群聚类实现电力设备红外图像的异常检测。首先, 对红外图像进行 RGB值校正, 并将校正的 RGB值映射到 Lab空间, 获取聚类所需数据集。核猫群聚类方法中的每一只猫代表着一种聚类划分, 用聚类中心点的坐标来对猫的位置进行编码。利用搜寻模式和追踪模式对猫群中猫的位置进行更新, 采用核函数引导的相似性度量构造目标函数, 通过迭代优化获得电力设备红外图像的多层分割聚类结果, 最终发现电力设备中的异常发热区域。实验通过与 k-means、fuzzy c-means和传统猫群聚类进行定量对比, 结果表明, 所提方法多层分割效果更好, 具有更佳的异常检测能力。

Traditional anomaly detection methods based on clustering are not very effective for the multi segmentation of electrical equipment infrared images. In this paper, an anomaly detection method based on kernel cat swarm optimization clustering is proposed. First, the infrared image RGB values are corrected, which are then mapped to the lab space to obtain the required data set for clustering. Each cat in the kernel cat swarm optimization clustering represents a clustering division, and the cat position is encoded with the coordinates of the cluster centers. The search mode and trace mode are used to update the positions of the cats. The objective function is constructed by the similarity measurement induced by the kernel function. Through the iterative optimization method, the multi segmentation result is obtained and the abnormally heating area in the power equipment is found. In the experiments, the proposed method is compared with k-means, fuzzy c-means, and traditional cat swarm optimization clustering. The results show that the proposed method has better multi segmentation performance and better detection ability.