Research On The Method Of Determining The Hydrophobicity Level Of Insulator Based On Image Recognition And Processing

As an important component of the power system,composite insulators are affected by bad weather for a long time,which will lead to the decline of water repellency and bring hidden dangers to power transmission.Therefore,it is very important to judge the water repellency level.When artificial water repellency is detected,the number of pictures is too large,it takes a lot of time,and the recognition rate is not high.This paper combines cutting-edge image recognition and processing technology to automatically identify and determine the hydrophobicity level,with good efficiency and high accuracy.Firstly,in order to improve the contrast of the image,this paper preprocesses the hydrophobic image of the insulator,extracts the B-channel grayscale image,further optimizes the Retinex algorithm,introduces the weighting factor on the illumination component,and performs the root number processing on the overall image coefficient.Finally,based on the NSCT transform and the improved multi-scale Retinex algorithm,the image is enhanced and preprocessed to improve the image quality.Secondly,in order to perform better binarization of images,the image N1 of the image HC1,HC2,HC3 image N0 and HC4,HC5,HC6,HC7 has a large difference in image formation characteristics,using the same kind.Image segmentation algorithm can not perfectly segment all levels of images.This paper proposes to perform two classifications before making seven levels of judgment,introducing a deep convolutional neural network,training the network into a model,and the model can intelligently The hydrophobic images are divided into two categories N0 and N1,and the corresponding image segmentation algorithm is adopted for the identified categories to segment.If it is identified as N0 class,this paper uses an improved edge extraction algorithm based on inter-class matrix iterative double threshold and image morphology to segment the image.Firstly,the edge extraction algorithm is used to extract the edge of the insulator water droplet,and then use the image morphology.Learn to expand,improve the area filling and opening operation based on background seed points to segment the water droplets from the background;if it is identified as N1 class,this paper uses the threshold fusion image segmentation algorithm to binarize the segmentation,that is,use the Nick local threshold ofthe improved parameter k The image is binarized with the Ostu global threshold coefficient fusion algorithm.Finally,in order to accurately determine the water-repellent level,this paper extracts five feature quantities in the binarized image formed by image segmentation,and inputs the extracted feature sample data into the XGBoost integrated learning classifier for model training.The model was tested,and the total recognition rate of the grade judgment reached92.9%,which was higher than other classifiers.