Research On Insulator Location And Fault Detection Method Based On UAV Image

With the improvement of China ’ s aggregate national strength,the demand for electric energy for national development has increased steadily,and the construction and transformation of the national grid has also continued to move forward.The expansion of transmission network makes the task of periodic inspection and inspection of transmission lines more and more heavy.Intelligent UAV inspection technology has become a research hotspot because it can lighten a burden.Based on the images captured by UAV,this paper automatically identifies and locates the insulators and their fault parts in UAV images through image processing technology and deep learning algorithm.In view of the interference of image data in UAV images caused by blur,distortion and other factors on subsequent processing and recognition,UAV images need to be preprocessed.Firstly the image needs to be grayed,and then the noise reduction effects of mean,median and bilateral filtering methods on UAV images are compared.The results show that the median filter has the best performance in noise reduction.The traditional image processing technology is not ideal in extracting the insulator features from UAV images,and the detection method has low universality,this paper presents an insulator recognition and positioning method based on YOLOv4 algorithm.The algorithm optimizes the anchor frame determination method by optimizing the cluster center point,adjusts the neck network structure,and adds SPP module and CSP + module,and optimizes the loss function.The training effect of YOLOv4 algorithm and the improved algorithm are evaluated from training loss and validation loss.The test results prove that the improved algorithm has more stable detection performance.Finally,the detected images and their confidence,recall,accuracy and mean accuracy are compared.The mean accuracy of the improved algorithm in this paper is 2 %-4 % higher than that of the YOLOv4 algorithm.The experimental data confirm that the improved algorithm in this paper improves the accuracy,performs well in the detection of insulators in different environments,and has certain generalization ability.In view of the high difficulty and high error detection rate of direct identification of insulator fault in UAV images,a second-order detection method based on modified YOLOv4 algorithm is presented.Firstly,the insulator part in the image is identified,and then it is extracted as a new data set,and then the fault area of the insulator is identified.By comparing the confidence,recall,accuracy and average accuracy of the test results and the average accuracy of the four test sets formed by different processing methods,the validity and generalization capability of the second-order detection algorithm for insulator faults in this paper are verified.This paper has 47 figures,10 tables and 60 references.