Research On Visual Recognition Technology Of Power Distribution Equipment Based On Deep Learning

The power system is composed of power generation,transmission,transformation and distribution,in which the distribution equipment is directly oriented to users,and any change of its working conditions will have a direct impact on the safe operation of the power system and user experience.Therefore,it is of great practical significance to carry out the health monitoring of distribution equipment.Due to the variety and wide distribution of distribution equipment,patrol inspection has become the main means of monitoring the health status of distribution equipment.However,the current patrol inspection is mainly implemented by manual operation,that is,in the process of patrol inspection,the relevant personnel obtain the working status of distribution equipment with naked eyes or handheld devices,including the status of various components and the pointer readings of different meters It is not suitable for the rapid development of power system.In recent years,with the development of mobile robot and machine vision technology,the integration of the two has become a research hotspot,and has been gradually applied in the health monitoring of distribution equipment,reflecting the obvious technical advantages.However,the more extensive and effective application of this technology,in addition to the progress of mobile robot technology,depends on the development of visual detection and recognition technology,especially the organic integration of the latest research results of deep machine learning,to realize the visual detection and recognition of power distribution equipment working conditions,including all kinds of component states and different meter pointer readings,and endow it with strong robustness and high performance in complex environment interference Intelligent detection and recognition ability.Based on the above background,this paper proposes to carry out the research on visual recognition technology of distribution equipment based on deep learning.On the basis of understanding the research status and development trend of power distribution equipment inspection related technology,using deep machine learning and machine vision technology,this paper focuses on solving the key technology of power distribution equipment positioning and related meter recognition,and realizes the research and development of the system.At the same time,experimental research is carried out to verify the feasibility and effectiveness of the technology.In the first chapter,the significance of the research on the inspection and visual detection and recognition technology of distribution equipment is discussed.The research status and development trend of the related technology of visual detection and recognition of distribution equipment are summarized systematically.The existing problems and corresponding countermeasures of the current visual positioning of distribution equipment components and the visual recognition of meter reading are analyzed,and the research direction of the paper is defined.At the same time,the research contents and chapters of the paper are arranged.The second chapter is to establish the theoretical basis of deep learning.The theory of convolution neural network is analyzed,which lays a necessary theoretical foundation for the research of visual inspection and recognition technology of distribution equipment.At the same time,on the basis of clarifying the function and performance objectives of mobile robot system for power distribution equipment inspection,the overall scheme of robot inspection is designed,and the key technologies to be solved are summarized.In the third chapter,the visual positioning technology of distribution equipment components based on deep learning is studied.Based on the definition of YOLOv3 convolutional neural network model,combined with the advantages of YOLOv3 multi-scale detection,using the principle of network structure lightweight,simplifying the network structure,and based on the improved YOLOv3 algorithm pruning strategy,a distribution equipment inspection and calculation method based on improved YOLOv3 is developed,and the detection rate reaches more than 98%,which meets the practical application requirements of distribution equipment inspection.In the fourth chapter,the research of distribution equipment meter visual recognition technology based on deep learning is carried out.On the basis of defining the visual recognition framework of distribution equipment meter,YOLOv3 algorithm is used to locate and recognize the dial numbers on both sides of the pointer in the meter,and Mask-RCNN algorithm is used to segment the pointer image.At the same time,the contour of the segmented image is extracted to get the position of the two end points of the pointer in the image.By connecting the two end points,the coordinates of the intersection of the two end points and the digital line of the dial are determined,and then the reading of the pointer is calculated according to the angle ratio.In the fifth chapter,based on the theoretical and technical achievements of the above chapters,on the basis of completing the research and development of key software and hardware modules,the system integration is carried out,and a set of distribution equipment inspection system based on deep learning is developed.At the same time,the system is used for related experimental research to verify the feasibility and effectiveness of the technology and system developed in this paper.The sixth chapter summarizes the research content of the full text and looks forward to the further research in the future.