Superpixel-Based Improved Fuzzy C-Means Clustering for Electrical Equipment Infrared Image Segmentation

An improved super-pixel based FCM segmentation method for infrared image of power equipment is presented to solve the problems of low segmentation accuracy, slow convergence, poor selection of initial cluster centers and local optimization in traditional fuzzy C-mean (FCM) algorithm. First, a simple non-iterative clustering (SNIC) superpixel algorithm based on multi-feature fusion is used to pre-segment the image, and superpixels are used instead of pixels to express the image features, which reduces the subsequent processing complexity. Secondly, using the idea of maximizing the variance between classes, the gray value corresponding to the maximum value of gray histogram when the variance between classes is maximized is selected as the initial cluster center of the improved algorithm to avoid generating local optimal solution. Finally, combining the SNIC algorithm of multi-feature fusion with the FCM algorithm, the infrared image of power equipment is segmented. The experimental results show that the algorithm improves the under segmentation of the comparison algorithm on the contour of the device and the local high temperature area, improves the operation efficiency, and lays a foundation for the later fault diagnosis of power equipment.