Construction And Management Of Key Components And Defect Data Set Of Transmission And Transformation Lines

With the unprecedented development of deep learning and unmanned aerial vehicle technology,in the inspection of transmission line circuits,new inspection methods have emerged.The specific method is that the front end uses a camera mounted on the drone to patrol the line and shoot,and the back end uses target detection technology to identify defects in the shooting content.However,the recognition rate needs to be improved.Researchers mainly conduct research from two aspects: algorithms and data sets.Based on the perspective of the data set,the study found that in the published data set,only the Chinese power line insulator data set is related to the inspection content,but the data set contains only the insulator and the image of the self-explosive defect.The lack of data sets has brought great difficulties to research in this field.Therefore,this paper constructs and manages a set of key components and defect data sets for transmission and transformation lines based on deep learning.The main work is as follows:(1)A set of key components and defect data sets for transmission and transformation lines are constructed.In terms of images,through collating the data,it was found that the number of component defect images was small and the proportion of defects in the images was small.To this end,the Non-Maximum Suppression(NMS)algorithm is introduced,and the defect data amplification method based on the improved NMS algorithm is proposed and used.In terms of labeling,the labeling specifications were formulated and adopted,and the labeling work was completed.(2)The management specifications and methods for the data set in this paper are proposed.In terms of data set structure,in order to more efficiently use the data set to train the target detection model,the original data set structure,experimental data set structure and related management specifications were designed.In terms of management mode,two closed-loop management modes of "data-model-data" and "data-model-optimization model" are proposed to achieve mutual management of data and models.(3)Designed and implemented an image database management system.The system includes two parts: management data set and management model.Itis divided into six functional modules: creating data set,managing data set,model training,model verification,model detection and test results.The database is a My SQL database.The system simplifies the work of data set management and model training with a graphical interface,and further completes the management of the data set.