Intelligent Recognition Of Composite Insulators Hydrophobicity Grades Based On Image Processing And Machine Learning

Composite insulators are widely used in transmission lines due to their advantages such as small size,light weight,simple manufacturing process.However,as the operating time increases,its pollution flashover resistance continues to decline,which is intuitively manifested as the decrease in the hydrophobicity of the silicone rubber umbrella skirt.In order to ensure the stable operation of the power system,it is necessary to periodically check the hydrophobicity of the composite insulators,and remove the seriously aging insulators from the operation of the power system.The commonly used method of water repellency detection is the water spray classification method,but the traditional method of HC has strong subjectivity,in order to overcome this shortcoming,many scholars combine the water spray classification method with image processing technology.The water droplet/water mark characteristics in the spray image are extracted,and the relevant mathematical model is used to determine the hydrophobicity level.However,this technical method still has many shortcomings in practical applications.Therefore,in order to further study the determination of hydrophobicity of composite insulators and improve the accuracy of detection,the following research work has been carried out:(1)Perform grayscale,adaptive histogram equalization,and homomorphic filtering algorithms on the water spray image samples of composite insulators with various hydrophobicity levels obtained in the experiment to highlight the edge details of the water mark,and select the Canny operator edge detection and adaptive threshold segmentation algorithm,combined with morphological algorithm correction,to obtain water droplets or water trace binary images,and analyze the advantages and disadvantages of the algorithm.(2)A selection criterion of water spray image processing method based on the distribution interval of the consistency measure is proposed.According to the water spray image of the insulator under different working conditions,the texture consistency measure value of the image after the enhancement processing is obtained,and the best algorithm is selected accordingly.Using Blob analysis to extract the water-repellent characteristics of the binary image of water traces,and use it as input to construct a hydrophobicity classification model based on BP neural network,Bayes classifier and Support Vector Machines.The results show that the Support Vector Machines model is better than the other two models,The highest recognition accuracy can reach 97.1%.(3)Based on the classification of digital image processing machines,the deep convolutional neural network model and migration learning are combined for hydrophobic recognition.The original uncut water spray image is used as input,and the image is extracted layer by layer through convolutional layer,pooling layer,etc.Feature,use the insulator water spray sample data set to train the network.Finally,the determination of the hydrophobicity level of insulators based on deep learning was completed,with the highest recognition accuracy rate of 94.8%,which can realize online detection of the hydrophobicity of composite insulators.