Research On Defect Detection Method Of Insulators In Aerial Overhead Transmission Lines

With the expansion of power grid scale,the length of overhead transmission lines is increasing,and the workload of power grid inspection and maintenance is also increasing.Unmanned aerial vehicles(uavs)are widely used in power grid inspection.Image processing technology can be used to conduct all-round inspection of overhead transmission lines,which is of great help to timely find hidden safety hazards in lines and determine defects of components.Insulator,as an important part of power system,adopts the traditional detection method of artificial eye,which has low detection efficiency and high detection rate.Therefore,this paper studies the defect detection of insulator.Defect detection based on morphology of insulator is first,secondly the way positioning before testing was used to study joint defect detection algorithm,respectively based on the characteristics of double Adaboost algorithm and the algorithm based on improved Faster-R-CNN positioning of insulator is studied,based on residual neural network(Res Net-50)algorithm for defect detection after positioning the insulator.The main research contents of this paper are as follows:1.Based on morphological insulator defect detection,the original image is first converted from RGB color space to HSV color space,and threshold segmentation is conducted by using the maximum category variance method.Then,complete and clean insulator string is extracted by morphological processing.Finally,the position of the insulator is determined by straight line fitting and the model is established based on this,and the defect of the insulator is judged according to the number of pixels in the corresponding position of the insulator image.2.The insulator location of Ada Boost classification based on double features.The Ada Boost classifier was trained by extracting Haar features and HOG features,and three optimization strategies were adopted for HOG features including variable block size,dimension reduction of Fisher criterion and fast calculation of integral histogram.The experimental results show that the method can locate the insulator with only 564 weak classifiers,and the error detection rate of the final algorithm is 8.3% and the miss detection rate is 12.4%.3.Based on improved Faster R-CNN insulator localization algorithm,this algorithm through the multilayer characteristics of convolution integration,improve the algorithm of target feature extraction ability,improve the proportion of the anchor box,and decrease the size of the anchor box so as to enhance the capacity,the recognition of small scale insulator to join in the process of training the multi-scale negative sample training and difficult mining strategy,through the training sample data of 30000 copies,eventually the mean average precision(m AP)of 93.6% than the original algorithm Faster-R-6.8% higher than that of CNN.4.The above two localization algorithms are combined with the residual neural network(resnet-50)defect detection algorithm respectively.Finally,it is concluded that the accuracy of the joint defect detection algorithm based on the improved fast r-cnn algorithm is 16% higher than that based on the dual feature Ada Boost algorithm and the overall time of joint detection is reduced by 0.76 s.