Research On Insulator Fault Detection Algorithm Based On Deep Convolution Neural Network

As one of the most common and break down easily components of transmission lines,insulators play an important role in transmission lines.It has great significance for maintaining the safety operation of transmission lines that finding fault insulator in time.UAV has become the trend of power line inspection instead of manual inspection.Automatic identify insulator fault from big data of aerial insulator images is a frontier interdisciplinary topic.This paper first analyzes the characteristics of aerial insulator images,and points out the characteristics of the images.Such as complex background,unstable quality,different viewing angles,different pixel sizes,and different positions of target in the image.These characteristics has brought great challenges for fault recognition method based on image processingAiming at the characteristics of aerial insulator images,an insulator recognition and self-shattering fault image recognition algorithm based on deep convolution neural network is proposed.The algorithm cuts the original image into several sub-images,and uses convolution neural network(CNN)to train the sub-images.By recognizing the sub-images,the different patterns of original images can be recognized.A training sample library containing 1220 sub-graphs with size 48×48 was constructed by using 64 original images.9 layers CNN was trained with training sample for insulator image recognition,and the training accuracy was 98.52%.The trained CNN model was tested using 151 original images with a correct rate of 98.01%.Collecting 2610 sub-graphs from 205 original images to establish a training sample database for insulator fault image recognition was trained on the construction of 9 layers CNN.The correct rate of training is 99.5%.The trained CNN was tested with 341 original images with a correct rate of 98.53%.As a result,the proposed algorithm can effectively identify the insulator image and insulator fault image,and has strong generalization ability.A self-shattering insulator segmentation algorithm combined DCNN and Otsu is proposed on the basis of identification of self-shattering insulator images.This algorithm uses the DCNN recognition results to optimize the Otsu segmentation results.The experiment shows that this method can effectively remove the tower pole background which is difficult to remove in the general segmentation method.An insulator fault detection system based on deep convolution neural network is preliminarily developed,including insulator image recognition and self-shattering insulator image recognition,which can provide certain technical support for insulator fault detection.