Research On Insulator Fault Detection Method Based On Images

In the construction and development of transmission lines,the reliability management of equipment and components is one of the basic and key tasks to eliminate and avoid accidents.It mainly includes condition monitoring,fault early warning,fault detection and troubleshooting of components such as insulators,ground conductors and metal equipment.Insulators often have multiple faults,and the amount of insulator image data is increasing.So,it is of practical significance to study the detection of insulator images and their faults.At present,most methods for insulator fault detection are to mark the insulator and its fault location,and do not express key information such as the attributes of the component,the type of fault and the degree of fault.Therefore,combined with the actual needs in the application,a systematic image-based insulator fault detection method is proposed to realize multi-modal conversion from image information to text information,which provides an algorithm basis for more efficient and rapid completion of insulator troubleshooting and image retrieval.Firstly,it introduces the research status and method analysis of insulator target detection,fault detection and description in recent years at home and abroad,and then proposes three aspects of research content and designs the overall scheme.In the insulator image data preprocessing stage,a collaborative labeling algorithm that effectively improves the labeling efficiency is proposed to obtain the classification and description labels of the image,and the deep learning YOLOv3 model is combined with the traditional color preselection model to achieve the insulator target extraction.In the insulator image classification model training method,two different models of VGGNet and Res Net are used to conduct multiple groups of comparative studies.The three key attribute information of the extracted insulator image string number,material and fault are classified.The comprehensive results prove that the method based on Res Net has higher accuracy and better generalization ability.Finally,in the insulator image and fault description method,a template-based insulator image description algorithm is proposed.The two classification networks obtained previously are respectively integrated with the description algorithm to form two complete image description models,and then the readable sentence description of the insulator image is output,including the description of the insulator fault.The results show that the description of insulator images can be effectively obtained by both methods,and the model constructed based on Res Net performs a lower error rate.