Research On Detection Method Of Aerial Transmission Line Insulator Based On R-FCN And Knowledge Expression

Insulators are a large number of components in transmission lines.Due to long-term operation outdoors,their surface is easily damaged,which causes serious operational safety problems for the entire line.Insulators in the images can be detected quickly and accurately by means of computer vision.It can greatly improve the efficiency of maintenance which is important to ensure the safe and effective operation of power grid.In this paper,the aerial transmission line insulator target detection task based on Region-based Fully Convolutional Networks(R-FCN)is divided into two stages: the detection of insulator target in aerial transmission line image and the detection of insulator defect target in aerial transmission line.The existing detection methods for insulator target cannot meet the high accuracy and high speed requirements of large-scale aerial transmission line insulator detection tasks.An insulator target detection method based on modified R-FCN in aerial inspection image is proposed.Firstly,according to the aspect ratio feature of insulator targets,the aspect ratios of proposals in the R-FCN model are modified to 1:4,1:2,1:1,2:1,4:1,and the scale of proposals are modified to 64,128,256,512.Then,in view of the occlusion problem in insulator image,an Adversarial Spatial Dropout Network(ASDN)layer is introduced into the R-FCN model to generate the samples of incomplete target feature by masking part of feature map,which can improve the detection performance of the model for samples with poor target feature.The average detection rate of R-FCN model reaches 77.27% in the data set containing 7433 insulator targets.The average detection rate of the modified R-FCN detection method is 84.29%,which improves 7.02%,and the detection frame is more suitable for the target.In the object of the insulator defect detection in the aerial transmission line image,the existing detection methods are all for a certain type of defect,and the problem of processing is relatively simple.Firstly,According to the knowledge that the feature of different types of insulator defect categories in the depth model has a certain similarity,classifies many insulator defect types.And an image data set of insulator target defects in aerial transmission line is established,which includes three types of insulator defects: insulator bunch-drop,insulator damage and insulator pollution.And based on this data set,the R-FCN model is fine-tuned.Then,the hyper parameters of the depth model are adjusted,and the average accuracy obtained by the fine-tuned R-FCN model is up to: 43.35%.The detection results of insulator defect targets in aerial transmission line insulator images are improved obviously.