Research On Recognition Method Of Transmission Line Icing Based On Vision And Design Of Recognition System

Affected by the air temperature,complicated terrain,and various weather conditions,the supercooled water droplets attached to the power line will eventually form ice due to the release of latent heat.The icing of transmission lines seriously affects the safe supply of electricity.Therefore,monitoring the icing conditions and measuring the icing thickness of transmission lines have become important research topics for many scientific research institutes.Traditional methods of detecting ice on power transmission lines include calculation methods based on mechanical models,simulated wires,and the like.The method based on the mechanical model can calculate the equivalent thickness of icing on the power line,but cannot obtain the local thickness of the icing on the power line.The simulated wire method can only indirectly measure the icing thickness.The measurement results are compared with the actual transmission with high voltage and high current.Line icing is different.The method based on visual measurement of ice thickness has the advantages of high accuracy and low cost.Therefore,this article chooses a visual method to realize the identification of ice coating and the measurement of thickness.The main contents of this article are as follows:(1)Research on ice processing recognition method based on image processing.Firstly,this paper selects bilateral filtering,histogram equalization,and adaptive mean filtering algorithms to preprocess the ice-covered image.After preprocessing,it uses Canny algorithm to perform edge detection to reduce false edges in the image.Secondly,first roughly divide the power line area,and then use Hough straight line to detect the ice-covered edges.Experiments show that this simple method can effectively reduce the interference of the image background and improve the accuracy of ice-covered edge recognition;Finally,a straight line detection method combining Hough transform and least squares method is proposed.Firstly,the centerline of the ice-covered edge is fitted.The centerline is used as the judgment condition to fit the upper and lower edge straight lines to complete the detection of ice on the power line.(2)Research on ice recognition method based on deep learning.It is mainly to improve the Seg Net network model widely used in the field of semantic segmentation.Firstly,a lightweight convolutional neural network Mobile Net was proposed to replace VGG16 as the coding structure of the Seg Net network,and the number of parameters was reduced from the original 14.71 million to 1.6 million.Secondly,in the decoding part,the skip structure of FCN is used for reference,and the position information of low-level features is integrated to improve the accuracy of the final segmentation.Up-sampling is achieved using bilinear interpolation.Finally,a detection experiment was performed on the ice-covered data set of the transmission line constructed in this paper.The results show that the MIo U value of the improved Seg Net network and the Seg Net network proposed in this paper increased from 78.2%to 81.3,an increase of 3.1 percentage points.In addition,the model sizes of the Seg Net network and the improved Seg Net are 45.19 MB and 24.74 MB,respectively.The size of the Seg Net network model based on Mobile Net is only 21.86 MB.Among these three algorithms,it can run on the mobile embedded platform jetson TX2.(3)Developed an ice monitoring system for transmission lines.This system integrates the improved Seg Net ice-covering detection algorithm proposed in this paper,and has functions such as ice-covering detection,thickness measurement,and data storage.The icing scene of the power transmission line was simulated under laboratory conditions,and the functions of the monitoring system were verified by experiments,which has certain application value.