Shadow Removal Hyperspectral Imaging for Outdoor Aging Evaluation of Composite Insulators

Composite insulators are widely used in power grids, and their aging with increased operational years is a key factor contributing to insulator flashovers and power outages. Accurately detecting the aging state of insulators is crucial for the stability of the power grid. Hyperspectral technology has been preliminarily applied to the state detection of insulators. However, under natural sunlight in actual operating environments, the shading caused by occlusion of insulator sheds results in significant reduction in detection accuracy. This has become the major obstacle in applying hyperspectral imaging for field applications of insulator aging levels. Therefore, this study proposes a shadow removal model based on improved color-lines (ICL) for ultraviolet aging assessment of composite insulators under natural light. First, accelerated UV-aged samples were prepared to extract spectral signatures from various aging levels, training a robust support vector machine (SVM) classifier. Next, the background pixels in the hyperspectral images of insulators are removed based on spectral differences between different materials. The shadow regions are located using color space conversion methods, followed by the progressive removal of both umbra and penumbra areas. Combined with a brightness map smoothed by Weighted Least Squares (WLS), the shadow-free hyperspectral images are generated. Finally, the trained model is used to classify the aging levels of the shadow-restored insulators, with the overall accuracy improving from 75.00% to 94.44%, outperforming existing shadow recovery algorithms. This method is of significant importance for field detection of insulator aging levels.