基于改进 PCM 聚类方法的绝缘子图像分割

针对绝缘子图像背景复杂多样而导致提取绝缘子区域困难的问题,提出了一种改进的可能性 C-均值聚类方法(PCM) 对绝缘子图像进行分割。 方法主要基于两个方面进行改进,一方面通过定义局部相关因子、引入图像的空间局部信息以增强对 噪声的抗干扰能力、提高分割精度;另一方面通过在损失函数中加入类中心相斥项缓解传统 PCM 聚类中心点重合问题。 实验 利用人工合成数据和复杂背景的绝缘子图像对比该算法与 FCM、PCM、K-means、KFCM 和 IFCM 算法的聚类分割性能。 结果表 明改进 PCM 对噪声抗干扰能力更强、聚类精度更高,且对绝缘子图像的平均分割误差为 0. 153,相比其他对比方法对复杂环境 下的绝缘子图片有更好的分割性能。

Segmentation method of insulator disk based on improved PCM

Aiming at the problem that the complex and diverse background in insulator images makes it difficult to extract insulator regions in practice, proposes an improved PCM clustering algorithm to segment the insulator image. To improve the algorithm from two aspects,firstly, by defining the local correlation factor, introducing the spatial local information of the image enhances the anti-jamming ability and improves the segmentation accuracy, secondly, adding the repulsive term of class center to the loss function alleviates the central point overlapping problem of traditional PCM. In the experiment, using artificial data and images of insulator with complex background to compare the proposed algorithm with FCM, PCM, K-means, KFCM and IFCM clustering algorithm. The results show that, the improved PCM has stronger anti-interference ability and higher clustering accuracy, which has better segmentation performance for insulator images than other contrast methods, and the average segmentation error is 0. 153.