Research On Insulator String Detection,Segmentation And Self-Explosion Fault Identification Method In Aerial Images

The insulator string was a key device in the high-voltage transmission line.It was mainly used to support and fix the current-carrying conductor,prevent the current from returning to the ground,and was exposed to the outside for a long time.It was prone to failure and was the monitoring object in the power line.In this paper,the aerial image in power line monitoring was taken as the research object.The deep learning algorithm was used to realize the identification and segmentation of the insulator string in the aerial image,and the self-explosion fault identification and localization of the identified and segmented insulator strings were performed.The main research work was as follows:(1)In the identification process of the insulator strings in the aerial image,based on Faster R-CNN,the region of interest generated in the regional suggestion network(RPN)was classified,and then the bounding box coordinates were corrected by the bounding box regression to realize the identification of the insulator string;for the identified insulator string,it was fine-tuned based on the full convolutional neural network(FCN),making it more specific to the dataset that needed to be segmented.The experimental results showed that compared with the existing methods,the proposed method could realize the identification and segmentation of insulator strings under different illumination conditions,different shooting angles and complex background interference,having high precision and robustness.(2)In order to realize the identification and localization of the self-explosion fault of the insulator string,an improved template matching algorithm was firstly established,and a single insulator template library was established.OpenCV and its matchTemplate function and minMaxLoc function were used to correct the image orientation while achieving the best matching of the template,and then calculated the pixel distance matching the adjacent two insulators,and located the self-explosion fault.In order to realize the visualization of self-explosion fault identification and positioning,improve the speed and accuracy of recognition and positioning,the deep learning algorithm was used to improve Faster R-CNN.By the deconvolution layer added,the anchor point frame initialized and the multi-feature layer information fusion,the influence of insufficient feature quantity and overgeneralization could be eliminated,and the algorithm was made more specific to insulator dataset.The experimental results show the effectiveness of the two improved methods.(3)The identification and segmentation of insulator strings,and the identification and location of self-explosion fault platform was built,which used high-performance computing servers as hardware and Tensorflow and Caffe as software environments.Various algorithms were tested on the platform,the performance of different algorithms was analyzed,and the advantages and disadvantages of the algorithm were evaluated.The research in this paper had certain reference value for the identification and segmentation of insulator strings and the identification and location of self-explosion faults in aerial images.