Research On Image Segmentation Method Of Composite Insulator Hydrophobicity Based On Cloud Model

As one of the most vital safety equipment to ensure the safe operation of trains,roof insulators are exposed to extreme environments such as wind and sand for a long period.When the concentrations of sand and dust are too high,they can break through the air gap between the umbrella skirts and cause a fouling accident,resulting in voltage fluctuations or even blackouts in the supply and distribution system,posing safety hazards and financial losses to transportation.It is critical to investigate the mechanism of composite insulator performance evolution in order to reduce the incidence of fouling and flashing occurrences.To evaluate composite insulator performance,umbrella pressure resistance,umbrella voltage distribution,water repellency,and other metrics are employed.This thesis focuses on the image-based hydrophobic detection methodology in order to improve the accuracy of hydrophobic grade distinction,the key research topic is as follows:1.To improve the accuracy of water repellent class classification,a transition cloud model with higher edge extraction capability is developed,solving the problem that the traditional cloud model is ambiguous in identifying nearby level of insulators.Firstly,above0.5 conviction,the transition cloud model is constructed using transition state theory and the steps of rounding-mirroring-stretching-regeneration of cloud droplets.The image-based insulator hydrophobicity class determination methodology and the classic cloud model are contrasted on this basis.The results show that,provided the uncertainty of water droplet edge information is guaranteed,the proposed method can identify the hydrophobicity class of insulators at the junction more accurately than previous techniques,giving a superior technological solution.2.To solve the issue of excessive histogram peaks in water-repellent images,which has a substantial influence on segmentation iteration efficiency and accuracy,presented an improved Gaussian mixture model cloud transformation strategy.K-means++ is used to optimize the Gaussian mixture model’s a priori starting values by selecting and refining the initial granularity of image information.As experimental samples,chosen two insulators with varying degrees of damage in real-world settings,with the Gaussian cloud transformation methodology used for simulation before and after improvement.When segmenting insulator pictures with more scratches in this type of harsh environment,the improved approach in this part takes roughly twice as long as the usual Gaussian cloud transform,and the misclassification rate and quantity of misclassified water droplets are reduced.3.In order to improve picture processing time,the spectral clustering image segmentation methodology is applied to picture segmentation for the problem of recognizing the hydrophobic grade of insulators with good hydrophobicity and general conditions.The water-repellent image of insulators is first subjected to a number of pre-processing stages,including grayscale,histogram equalization,and filtering,with the filtering results quantified.The processed images are then used as input,and three sets of insulator images with varying water repellency are simulated and compared using k-means,fuzzy c-means,and spectral clustering,in that order.Finally,the discontinuous edges are improved for batch grade assessment employing morphological filling corrosion and other processes.