Research On Crack Detection Model And Total Analysis Method Of Insulator Crack Based On Computer Vision

Insulators are important part of transmission and distribution lines,which are widely used in power systems,with a large number.Because Insulators work in complex and harsh environment for a long time,cracks are easy to appear,causing grid accidents and economic losses,which seriously affect the safe operation of power system.Therefore,in the inspection process,it is very important to find Insulator cracks in time and report relevant information accurately.However,with the rapid development of power grid construction,facing the huge power grid that has been built,the traditional Insulator crack detection method has high cost,low efficiency and high rate of missing detection,which can no longer meet the increasing demand of power safety.In recent years,unmanned aerial vehicles and other intelligent devices have brought technological innovation to power detection,while deep learning has given power to intelligent detection devices and upgraded them.In this paper,deep learning algorithm is used to study Insulator crack detection and crack total analysis in UAV detection images,which plays an important role in realizing the intelligence and accuracy of Insulator crack detection.The main research contents and innovations of this thesis are as follows.(1)In order to solve the problem of lack of sample data of Insulator crack label,an Insulator crack data set is constructed.Firstly,the Insulator crack images are collected and analyzed,and the image are processed according to their features.Secondly,the processed images are labeled,and the data sets are generated and enhanced.(2)Aiming at the problems that cracks are easy to be lost and isolated noise is difficult to be eliminated in the existing crack detection algorithms,a crack detection network based on reversible pyramid and balanced attention is proposed.Firstly,a reversible pyramid model is proposed,in which feature pyramid and improved inverted feature pyramid are introduced between encoder and decoder to deepen the fusion of global features and detail features,and thus improving the performance of fine crack detection;Secondly,in the decoding stage,the balanced attention module is introduced,and the balanced features are taken as guiding information to effectively eliminate isolated noise.Finally,in the learning stage,Focal Loss is selected as the loss function to control the weight of positive and negative samples in training,so that the model pays more attention to crack samples.Experiments show that the crack detection method designed in this paper can realize semantic segmentation with higher accuracy.Compared with other benchmark methods,the detection effect on small cracks is better,the isolated noise is effectively eliminated,and Io U in test set can reach 61.42%.(3)Aiming at the problems of difficult measuring direction and low precision of edge location of existing crack total amount analysis algorithm,a method for analyzing total amount of cracks in porcelain bottles based on improved skeleton location and edge distance measurement is proposed.Firstly,the crack detection result graph is processed,the crack skeleton and crack edge are extracted,the skeleton generation algorithm is optimized,the information loss in total analysis is reduced,and the key information is provided for total crack analysis.Then,on the basis of optimizing the skeleton,the information of crack length is obtained by distance measurement algorithm;Finally,the direction determination algorithm based on skeleton is introduced.The normal position is determined by calculating the measurement direction based on skeleton,and the normal distance between edges is measured to obtain the crack width information.By verifying and testing the self-built Insulation Crack data set of transmission and transformation line,the experiment show that compared with other benchmark methods,SLERA improves the accuracy of crack quantification,effectively avoids the influence of complex edges and their noises on total crack analysis,and can realize high-precision total crack analysis.On the Insulation Crack data set,the relative error rates of the length measurement results and width measurement results are less than 30%,which shows the effectiveness and robustness of SLERA.