Research On Insulator Fault Recognition Method In Power Inspection Image

The good working state of insulator plays an important role in the safe and stable operation of power system.In general,the insulator failure rate is high due to the relatively harsh environment of the transmission line.Failure to find the faulty insulator in the transmission line in time will bring significant security risks to the normal operation of the transmission line,so regular inspection of the insulator in the transmission line is particularly important.In recent years,it has become a research hotspot to realize the automatic identification of insulator fault by using power inspection image combined with computer technology.In order to improve the accuracy of insulator fault identification in power inspection,the algorithm of insulator fault identification based on image processing and deep learning was improved in this paper.The main work is as follows:Aiming at the problem of noise and lack of light in the process of power inspection image acquisition,this paper analyzes the types of noise in the inspection image,and selects the bilateral filtering algorithm to remove noise with good effects.At the same time,compared the results of various common image enhancement algorithms,MSRCR algorithm which is more suitable for power inspection image enhancement processing is adopted.The experimental results show that the insulator target information is more prominent in the power inspection image after image preprocessing,which is conducive to the smooth development of subsequent research on insulator positioning and fault identification.Aiming at the problems of long time consumption and low accuracy of image processing based insulator self-detonation identification method,an insulator self-detonation identification method based on spatial characteristics was proposed.The method realizes the segmentation,positioning and rotation of insulators in the inspection images by establishing the RGB color discrimination model and the minimum bounding rectangle method.In the horizontal positioning region of insulator,according to the difference between the normal insulator and the fault insulator in the space domain feature projection curve,the distance discriminant is established to determine the insulator self-detonation,which can accurately detect the insulator self-detonation region.Experimental results show that this method can accurately identify and locate self-detonating insulators in complex background,with a recognition accuracy of 90.2%,and a detection speed of 386 ms per single image of power inspection.Compared with similar methods,this method has higher identification accuracy and less time,so it has certain practical value.Aiming at the problem that the current deep learning based insulator self-explosion fault identification accuracy is insufficient,an improved YOLOv3 network structure insulator self-explosion detection method is proposed.Based on the idea of expansion convolution,the 16-fold down sampling unit in the YOLOv3 backbone network was improved to integrate more target information while ensuring the resolution of the convolution network.Meanwhile,by improving the distance measurement formula in the traditional K-means clustering algorithm,the size of the anchor box is more suitable for insulator self-burst detection.The experimental results show that the improved YOLOv3 detection network structure can improve the recall rate from 91.62% to 95.84% and the recognition accuracy from 96.9% to 98.4% on the premise of guaranteeing the real-time performance.In order to simplify the operation steps in practical application,the improved network structure combined with Py Qt5 development tool was used to complete the GUI interface of insulator self-explosion fault identification system,and the corresponding test was carried out.