Research On Co-segmentation Method For Aerial Insulator Image Based On Super-pixel

The aerial insulator images are characterized by complex background, low resolution, large in number and more fake target, et cetera. The traditional segmentation method causes user’s fatigue and results in bad segmentation quality. The co-segmentation method can segment common object in multiple images which utilizes the relationship between images that reduces user’s workload and improve the segmentation quality. This paper introduces a co-segmentation method for aerial insulator images based on thermodynamic anisotropic diffusion model, and proposes a co-segmentation method for aerial insulator images based on C-V model of automatic initialize contour with Hough detection and inpainting,and proposes a co-segmentation method for aerial insulator images based on co-random walks model.This paper firstly designs a co-segmentation method for aerial insulator images based on thermodynamic anisotropic diffusion model. The method firstly removes the text in aerial images and over-segments the preprocessed images into super-pixels, which is in order to achieve more accurate and fast segmentation. According to the thermodynamic anisotropic diffusion theory and the constructed graph network, we cluster all the super-pixels into the assigned numbers and extract the corresponding largest relevant region by temperature maximization among the images as the common insulator objects. The experimental results show that the proposed method can obtain good results which are instrumental to insulators’ recognition and fault diagnosis.To solve the problem of using the co-segmentation method for aerial insulator images based on thermodynamic anisotropic diffusion model could not segment insulator with adhesional condition, this paper proposes a co-segmentation method for aerial insulator images based on C-V model of automatic initialize contour with Hough detection and inpainting. The method firstly removes the text and uses Hough detection and inpainting algorithm to solve conglutination problem in aerial images. Then over-segment the recovered images into super-pixels by SLIC for fast segmentation. According to the GHT coarse localization, we select the initial contour of the C-V model and use the C-V model for images co-segmentation. The experimental results show that the accuracy rate of segmentation for the proposed method is higher than those for other methods. The method has good performance of automation, it can separate the target and background region and the fake target can be eliminated efficiently, which is instrumental to unmanned aerial vehiclel insulators’ state detection and fault diagnosis.Finally, to solve the problem of using the classical random walk algorithm causes error segmentation easily, and the classical random walk algorithm requires massive interaction to segment multiple images. Alternatively, there is no need for the time consuming iterative process of C-V model, this paper proposes an automatic co-segmentation method for aerial insulator video images based on co-random walks. Constructing collaborative graph network based on the relationship of the pixel feature of intra-image and inter-image, and obtaining corresponding seed points to reach the segmentation of multiple images precisely. We use the proposed method to complete aerial insulator image segmentation experiments. The results verify that the co-random walks co-segmentation algorithm has higher robustness compared with the other segmentation methods.