Detection And Fault Identification Based On Deep Learning In Aerial Transmission Line Inspection Images

With the rapid development of the power system industry,the amount of component damage in the transmission line is also increasing.In the past,power companies usually used humans to visually inspect transmission lines for maintenance and inspection.To reduce costs and potential threats to infrastructure,some companies have turned to unmanned aerial systems.Nowadays,many methods of automatic detection of potential faults have been developed,and the focus is on detecting small faults at the component level to prevent large faults in the power distribution system.These detections are usually done by automatically checking component status and partial failures before larger failures occur.Therefore,the use of UAV aerial inspection images to realize automatic detection and fault identification of insulators,replacing the traditional inefficient manual inspection methods to ensure the safety of power transmission systems,has important research and application prospects.In order to maximize the accuracy of detecting explosive insulators using drone images,this article first uses a two-stage target detector based on Faster RCNN to solve the problem of class imbalance in the training phase of the detector.Transmission lines taken from drones Locate the position of the insulator in the inspection image.After obtaining the position information,cut the insulator target along the detection frame and input it into the segmentation model for semantic segmentation.Finally,use the pre-trained insulator state classification model to determine the state of the insulator after segmentation So as to achieve the purpose of improving the accuracy of insulator detection and the accuracy of fault identification.Specifically,the main work of this paper is to enhance the insulator data set to improve the robustness of the model.At the same time,it analyzes and compares the algorithm processes of the classic detection,segmentation,and classification models.According to the image characteristics of the aerial insulator data set,combined with the focus loss function and the idea of difficult case mining,an insulator detection model that solves the imbalance problem is realized,and then combined with the insulator image characteristics after detection to design a segmentation model to enhance the segmentation effect of its boundary part,Combined with the image classification model,finally an insulator fault recognition model based on deep learning is obtained.Using a 5-fold random cross-validation strategy to complete the detection training and performance testing,the insulator state automatic recognition model based on deep learning constructed in this paper achieved the highest detection accuracy of 95.8% in the comparison experiment,which verified the model’s ability in insulator detection and failure Effectiveness in recognition.