| Power line inspection is crucial to the normal operation of the power line system, since it can discover the defects existing in the power line corridor as early as possible and avoid the happening of serious electrical accidents. Power line inspection by using helicopter is one of the components of the Smart State Grid, which can reduce the workload of the conventional manual power line inspection, decrease the corresponding risk, and improve the efficiency and accuracy of the power line inspection. In order to further liberate the manual works from the visual recognition of the huge power line images, many image processing techniques for aerial power lines have came into being with the rapid development of artifical intelligence and remote sensing technology. In particular, the technology of edge detection is an important tool for the recognition of power line components and the identification of power line defects. Firstly, the study of edge detection is of great theoretical significance, since there is no general edge detection algorithm. Most of the existing edge detection algorithms are for particular images. Secondly, edge detection algorithm is crucial to improve the intelligent level of power line inspection by helicopters. The performance of edge detection directly affects the recognition of power line components and the diagnosis of power line defects. Accurate and efficient edge detection provides reliable prerequisite for the recognition of power line components and the diagnosis of power line defects.Due to the complexity of aerial power line images, the higher noise level, and the low contrast between the power line and background, it is very difficult to obtain accurate edge detection results by conventional edge detection methods. Therefore, further research is still needed. In this paper, we mainly focus on the edge detection of aerial power line images, and put forward four innovative edge detection algorithms. The main contributions in this thesis can be summarized as follows:1. The conventional edge detection methods employ only the intensity gradient to extract object edges. They tend to miss the weak edges, representing the lower denoising ability when used to process the power lines images with complex background. In order to solve the problem, this thesis proposes a new edge detection method based on textural gradient. After the textural analysis for lots of power line images, we found that there exists obvious textural difference between the power lines and background. Considering the characteristics of power lines, we constructs a new operator on the basis of textural gradient to determine the candidate edge points. And then, inspired by the conventional steps of the edge detection methods, i.e., threshold denoising and edge linking, a new edge detection algorithm is proposed. Experimental results demonstrate that the proposed method obtains satisfactory edge detection results for the power lines with low contrast, preserving the bent information of the power lines at the point of broken-strands.2. The performances of edge detection and edge localization are always a pair of contradiction in conventional edge detection methods, e.g. Canny algorithm. In particular, it is more obvious when such method is used to process the power line images with complex backgroud. In Canny method, the Gaussian function is used to approximate the optimal edge detector in the sense of the famous Canny criterion. However, the pair of contradiction still exsits although it is balanced by the Gaussian scale parameter σ to some extent. In order to further address the pair of contradiction, considering the optimal geometry property of the Gamma probability density function (PDF), this paper firstly complements its definition at the origin and introduces the corrected version as the kernel function of the Canny edge detector. Meanwhile, an edge preserving parameter ε is added to make the pair of contradiction to be adjustable independently. With the Gaussian kernel function substituted by the corrected Gamma PDF, an improved edge detection algorithm is proposed. The advantage of the proposed algorithm has been validated by theoretical analysis and experiments on lots of aerial power line images.3. The existing active contour models focusing on segmenting textural objects tend to be trapped into local minima during the process of contour evolution, and they are sensitive to the position of the initial contour, lower convergence speed due to the numerical minimization algorithm on the basis of Euler-Lagrange and Gradient Descent Flow methods. In order to partition the complex aerial insulator images into sub-regions with closed smooth contours, this thesis proposed a Global Minimization active contour model with the Texture Descriptor (GMTD) by making full use of the edge information and texture information. In the proposed GMTD model, the Gray Level Co-occurrence Matrix (GLCM) is firstly extracted to describe the texture features of insulators and is calculated by the rapid Gray Level Co-occurrence Integrated Algorithm (GLCIA). The extracted texture features are devided into two categories:one with the stronger discriminative ability and the other with weaker ability. The second category is optimized by Principal Component Analysis (PCA) to better distinguish the different texture objects with low contrast. The proposed GMTD model can avoid the existence of local minima. A fast dual formulation is introduced for the efficient evolution of the contour. The experimental results on synthetic and real aerial insulator remote sensing images have shown that the proposed algorithm obtains more satisfactory segmentation results compared to the classical models in terms of accuracy, efficiency and independence of initial contour.4. Most of the real texture images represent the texture inhomogeneity to some extent. Compared with the homogeneous textures, the different texture regions of the texture object with texture inhomogeneity tend to be segmented into different parts. Focusing on this problem, this thesis constructs an energy function by using the difference of texture feature distribution inside and outside the evolving contour to drive the contour evolution. On the basis of the energy function, this thesis further puts forward an active contour model for segmenting the texture object with texture inhomogeneity. Meanwhile, the convexation of the proposed model is performed in the GMAC framework, and the minimization of the proposed model is solved by using the variational dual form technology, such that the model can converge to the global minimum quickly. Experimental results demonstrate that the proposed model can successfully segment the texture inhomogeneous insulators from the aerial power line images with complex background, obtaining satisfactory segmentation results and overcoming the drawbacks of the conventional active contour models. |
PDF Full Text Request