Research On Positioning And State Detection Of Insulators Of Transmission Lines Based On Computer Vision

Insulator is a kind of power device made of ceramic or glass,which can play the function of insulation and support in power transmission lines.China is vast in territory and the entire high-voltage transmission network is large in scale and covers a wide area,which has brought about a harsh transmission environment.Insulators are often left in the wild for long periods of time.In high pressure and complicated climate environment,once the insulator fails,the transmission will be unstable,and in severe cases,it may even cause a short circuit failure for the entire line.In the context of the rapid development of the power grid,traditional manual inspection methods are not only inefficient,but also costly,therefore,they cannot meet the requirements of the power grid in the current era.With the advancement of technology,especially the Internet of Things(IoTs)and computer vision,drones have been adopted in inspection of transmission line.With the rapid development of deep learning in recent years,it provides a feasible solution for the analysis of aerial photography inspection images of drones.Focusing on deep learning algorithms,in this article,the insulators in aerial images were researched for location recognition and status detection,and based on this,a system of line inspection by UAV was designed.The main research contents and research results in this article are as follows:(1)The neural network theory was introduced,and the advantages and disadvantages of common activation functions were analyzed.For the problem of many insulator strings in a single aerial image,an adaptive cropping algorithm was proposed.An image preprocessing algorithm was proposed based on the requirements of deep learning input samples.(2)An insulator positioning method based on the improved yolov3 algorithm was proposed.Darknet-53 was selected as the feature extraction layer.For the problem that the target of the insulator string has a small impact on the recognition rate,the number of channels for the feature target feature map of the third target detection was converted to 21.Based on the Tensorflow deep learning framework,the YOLO-V3 insulator detection network was trained and forward predicted.Through experiments,the intermediate layer feature map of the insulator sample image in this paper was drawn,which verifies the effectiveness of the improved localization algorithm.(3)A classification network of insulator states based on scale pyramid was propose.In this network,discrete wavelet transform was adopted to extract insulator features at different scales.Then,the images at different scales and the original image after wavelet transform were adopted to form different channels of data input,and a classification network was constructed to classify and judge them.The performance index of insulator state classification was defined,and the classification model in this paper was compared with the current mainstream neural network classifier model,which verified the effectiveness of the classification model in this paper.(4)The hardware platform of UAV intelligent line-tracking system was introduced.On OpenCV4.2 and Visual Studio2017 platforms,the above algorithm was ported to write software.The test video set was imported to verify the effectiveness of the algorithm.The experimental results showed that the positioning and status detection of the insulator can be implemented in the software,and the software can output the number of the pole where the insulator is located,which is convenient for the grid staff to exchange insulators,which improves the efficiency of transmission line inspection.