Research on infrared image segmentation technology of transmission equipment based on local area Medoidshift clustering algorithm

Abstract

The extraction of the fault area in the infrared image is the critical process in the intelligent identification of power equipment faults. Since the infrared image has the characteristics of low contrast, regional gray unevenness, and blurring, there exist great difficulties in fast and accurate image segmentation. To meet the demand of infrared image field processing for mobile inspection of power equipment, this paper proposes an algorithm for extracting faulty areas based on local area Medoidshift clustering. The method combines the characteristics of the thermal fault area and the grayscale adjustment mechanism for the similar pixels in the neighborhood so that the pixels in the fault area are clustered under the Medoidshift algorithm. At the same time, to speed up the clustering process, a neighborhood clustering method based on segmenting the entire image by iteratively computing the current target cluster mean is adopted. Experimental tests on the typical infrared images show that the proposed method is effective in region extraction. Compared with other methods, the method in this paper has better performance in the speed and accuracy of fault region extraction.