The Segmentation of Wind Turbine Defect Based on UAV Infrared Image

Abstract

During the operation of wind power systems, problems such as damage may occur, in order to keep the system running normally and safely, regular maintenance of wind turbines is essential. Currently, the main treatment is to detect the blades by carrying an infrared camera on board a UAV. However, infrared images can suffer from significant inhomogeneities due to the inconsistent response of the detector array of the infrared camera, the inconsistent spectral response of the infrared detector to environmental variations and incoming background radiation, and the combined effect of the camera's optical lens. The inhomogeneity of an image can greatly affect the accuracy of image segmentation. To solve this problem, In this paper, we propose a method for rapidly segmenting inhomogeneous images. Specifically, we introduce global and local information of the image into the energy term. We also introduce a detail map as the reference map for the edge stopping function, which improves the robustness of the image to the initial contour and improves the segmentation accuracy of inhomogeneous images while speeding up the segmentation. Correction of bias field is also realized simultaneously. The results demonstrate that our method achieves average accuracy of 96% in the task of performing segmentation of wind turbine defects with the shortest running time. Notably, our segmentation efficiency surpasses that of mainstream level-set techniques.