Research On Infrared Image Insulator Detection Method Based On Deep Learning

Transmission line inspection is an indispensable hub for ensuring safe and stable operation of the power grid.Insulators are important components of transmission lines.In the power system,defects and faults occur frequently,and on-time inspection is required to ensure the safety of the power system.As a popular inspection method,UAV inspection has obvious advantages such as fast inspection speed and high efficiency compared with traditional manual inspection.UAVs can capture a large number of inspection images by equipped with high-definition cameras.These inspection data are only analyzed and detected manually.The workload is huge and inefficient,and the experience and quality of the staff may affect the test results.Therefore,an automated test method is needed to diagnose the insulator.In the complex scenario of images taken by drones,the rapid diagnosis of insulator defects is a critical issue that needs to be addressed urgently.In this paper,we first select an algorithm based to detect insulators.Then,using the classification network,the detected insulators are divided into defective and non-defective types to identify defective insulator images.The following are the main work done in this paper and the experimental results obtained:(1)A collection of 3 types of commonly used insulators,a total of 1091 insulator image sample sets in the natural environment,and the insulators were calibrated.The insulator image was expanded by image rotation,translation,cropping,etc.,and the expanded sample was increased to 12,480.This data set can be commonly used in related research on insulator identification and image recognition in other fields.(2)The detection of insulators is realized by building Faster R-CNN,which solves the problem that the traditional detection algorithm is insufficient in robustness,weak in generalization ability and low in detection algorithm accuracy.Firstly,by studying the various parts and different uses of Faster R-CNN,according to the actual engineering needs and hardware configuration,the selection and design of the network structure of Faster R-CNN are realized,and a suitable insulator network detection model is constructed.Secondly,the extended sample is used as the training sample for network training,and combined with the relevant parameter-adjusting technology,the accurate insulator detection network model is finally obtained.(3)This paper uses CNN algorithm to classify and identify insulator defects to achieve the purpose of intelligent and fast diagnosis of insulator defects.Firstly,the cutting object is used to cut out the detection object,and the insulation is divided into two types: defective and non-defective.The well-constructed convolutional neural network model is used as the classification and identification network of insulator defects to classify and identify insulator defects,which has high recognition.The rate can be used instead of manual detection to reduce the error in the judgment caused by the inspection staff’s experience,so that the grid operation becomes more secure and stable.