Intelligent Fault Identification Of Insulators Based On UAV Vision

As an important component of the transmission line,the insulator plays an insulating role in a power system.Once the fault occurs,the power supply working process will be interrupted or even a large range of power failure will occur.Therefore,timely detecting insulator fault is of great significance for the normal operation of the whole power supply system.The fault identification of insulator images taken by UAVs is discussed in this paper.The main research works are as follows:(1)In order to solve the issue of insulator crack identification,we first preprocessed the insulator images and improved the quality of these images.Then,the Canny algorithm with a better effect is selected by comparing several common edge detection algorithms.To avoid the shortcomings of the traditional Canny algorithm,an improved Canny algorithm is proposed for insulator crack identification.The simulation results show that the improved Canny algorithm can detect the crack edge more clearly than the traditional Canny algorithm.(2)The traditional mathematical models for self-explosion fault identification of insulators only consider that the distance between each two adjacent insulators is equal in images.However,there are overlaps between insulators when UAVs take pictures of insulators from different shooting angles,thus the distance between adjacent non-self-detonation insulators may be different.To tackle this issue,this thesis improved the genetic algorithm and combine it with the optimal entropy threshold determination method(KSW entropy method)to split insulator.In addition,we use morphological processing to improve the segmentation effect for pure insulator string.We also proposed an improved insulator explosive fault identification algorithm which fits each insulator spacing based our designed linear equation.The proposed approach can make explosive fault identification of insulators have universality.(3)In order to make insulator self-explosion fault identification more intelligent,an insulator self-explosion fault identification method based on Faster R-CNN is proposed.Due to the actual insulator fault proportion is small and difficult to detect,we adopted the idea of identifying the insulator first and then identifying the fault.In the stages of insulator identification and insulator fault identification,we compared the performance of the AlexNet,VGG16 and VGG19 feature extractors.By analysing the experimental results,we concluded that VGG19 feature extractor achieved the best result in insulator identification stage and the performance of the VGG16 feature extractor is the best in insulator fault identification stage.