Research On Identification Method Of Power Components In UAV Inspection Images Of Transmission Lines

The integrity of power components is one of the preconditions for the normal operation of the whole transmission system.The inspection of high-voltage transmission lines is helpful to find all kinds of faults of power components in time,so as to ensure the stable operation of the power grid.The application of UAV in the inspection of transmission line makes the inspection process break through the limitation of terrain and improve the inspection efficiency.However,in the process of UAV inspection at present,the judgment and recognition of various faults are mainly carried out by manpower.Therefore,the automatic recognition of faults in UAV inspection images by using image processing technology is now a research hotspot,and the accurate recognition of power components in images is the premise for the automatic judgment and recognition of transmission line faults in inspection images.Therefore,this paper studies the identification method of power components in UAV inspection image of transmission line.The main research contents are as follows:In view of the noise problems existing in the process of UAV’s image shooting and image transmission,after analyzing the noise characteristics in the inspection image,a bilateral filtering algorithm with good denoising effect is used to denoise the inspection image of UAV;in view of the problem that some inspection images of UAV have insufficient brightness,MSRCR algorithm is selected to enhance the images with insufficient brightness.Through image denoising and enhancement,the information in the image is displayed more accurately,which lays a foundation for the accurate identification of power components in the following UAV inspection image.Aiming at the object of insulator string in high-voltage transmission line,a method of power component recognition in inspection image based on SIFT-BOW feature is proposed.Methods firstly,candidate regions were extracted by selective search,then SIFT feature vector was extracted,and SIFT feature of the whole image was transformed into a high-dimensional vector by word bag model.Finally,support vector machine(SVM)classifier was used to recognize the object.The experimental results show that the accuracy of identification of insulator samples by “SIFT-BOW + SVM” method is 89.3%;in the test of complete UAV inspection image,the AP value of “SIFT-BOW+selective search” method is 0.721,Recall value is 0.681,and the recognition time is 18.3 seconds per piece.Compared with the method of extracting candidate areas by sliding window,the recognition speed is greatly improved.The object detection algorithm based on deep learning is used to identify insulator string,corona ring and spacer.In the framework of Darknet deep learning,the algorithm of YOLOv3 is simplified by removing the large feature graph,and a two-scale YOLOv3 algorithm is obtained.In tensorflow deep learning framework,Fast R-CNN object detection algorithm is used to identify three kinds of power components in UAV patrol image.In Darknet deep learning framework,YOLOv3 object detection algorithm and two-scale YOLOv3 algorithm are used to identify three kinds of power components in UAV patrol image,and then the recognition effect of the three methods is compared.The experimental results show that: the mAP value of Faster R-CNN is 0.8354,the Recall rate is 0.88,and the recognition speed is 185.3ms/sheet;the mAP value of YOLOv3 is 0.879,the Recall rate is 0.87,and the recognition speed is 36.85ms/sheet;the mAP value of two-scale YOLOv3 is 0.863,the Recall rate is 0.86,and the recognition speed is 30.75ms/sheet.Using the GUI library Tkinter of Python language and dnn module of OpenCV,the GUI recognition interface is made and the recognition test is completed.Through experimental comparison,it can be found that the recognition method based on deep learning has higher accuracy and faster recognition speed in the recognition of power components.The research in this paper lays a solid foundation for the subsequent use of image processing technology for fault identification in transmission lines.