Research On Typical Fault Identification Method Of Transmission Lines Based On Object Detection And Instance Segmentation

Overhead transmission lines built in the wild are prone to faults caused by natural factors,posing safety hazards to the power system.Typical faults include insulator burst,grading ring falls off,and bird nest.Transmission line inspection based on drone aerial images can effectively troubleshoot faults and is safe and reliable,thus gradually gaining popularity.However,aerial images also have various problems,the most prominent of which are the complex background,small proportion of targets to be inspected,which leads to low fault identification accuracy,and high resolution,which leads to slow identification speed.Therefore,quickly and accurately identifying typical faults in aerial images has become a research hotspot in fault identification of transmission lines.Based on the improved object detection YOLOv5 s algorithm and instance segmentation SOLOv2 algorithm,this paper proposes a fault identification method that combines the above two algorithms to achieve accurate and fast identification of typical faults in transmission lines in aerial images.By collaborating with the power grid company to independently construct a transmission line dataset,the effectiveness of the proposed method is verified.The overall process of the method is divided into two stages.The first stage is to locate the area to be detected,and the second stage is to identify faults.The specific research content and methods are as follows:(1)In the first stage,in order to solve the problem of partial overlap of insulators,an improved YOLOv5 s object detection algorithm is proposed,utilizing DIo U＿NMS optimizes the box filtering of YOLOv5 s to improve positioning accuracy.Construct a transmission line dataset using data enhancement and train the improved YOLOv5 s for object detection,in order to quickly and accurately locate the area to be inspected.Then,the box coordinates obtained from object detection are cropped to separate the images inside the box and obtain the image of the area to be detected;(2)In the second stage,in order to solve the problem of difficult parameter regulation caused by the decoupled structure of SOLOV2,an improved SOLOV2 instance segmentation algorithm is proposed,which utilizes Rank&Sort Loss(RS Loss)to simplify training and avoid the performance impact caused by parameter regulation,making the training parameters more suitable for the transmission line dataset,thereby improving identification accuracy and speed.Finally,the improved SOLOV2 is trained using the image of the detected area obtained in the first stage,and instance segmentation is performed.Through dynamic convolution,a mask is generated globally to achieve accurate fault identification.The experimental results show that the identification method based on object detection and instance segmentation can effectively filter out complex backgrounds,increase the proportion of targets to be inspected,reduce resolution,and improve the accuracy and speed of fault recognition.In typical fault identification of transmission lines,compared to the original SOLOv2,the identification accuracy and speed are significantly improved,and the mask is more accurate.