Research On Remote Sensing Map-Based Power Transmission Tower Identification And Positioning System Based On Deep Learning

The demand for electricity for industries and households is rapidly increasing due to societal development and progress.In the first two months of 2023,there was a 3.2%YoY increase in the national industrial electricity consumption,reaching 857 billion kilowatt-hours.Ensuring safety and stability of the power grid is vital for the national economy.Therefore,regular inspections of transmission lines are crucial in maintaining power grid safety and stability.However,the efficiency in overhead line inspection and fault handling might be impaired due to untimely grid information updates,scattered storage of transmission line information and complex network structures.This thesis aims to mitigate these challenges by exploring the identification and positioning system of transmission towers through remote sensing maps using deep learning methodologies.The primary research areas of this thesis comprises of the following:The thesis begins by examining the research on satellite remote sensing data processing technology and target detection algorithms through experiments conducted on satellite remote sensing images.The research includes various processing methods such as radiation calibration,atmospheric correction,data registration,image fusion,linear stretching,and super-resolution.The thesis compares the performance of YOLOv5 and YOLOv3 algorithms using the open-source remote sensing data set DOTA.The mean average precision(mAP)of YOLOv5 and YOLOv3 algorithms were found to be 76.3%and 74.8%,respectively,with the corresponding F1 values of 76.33%and 75.54%,respectively.The research concludes that the YOLOv5 algorithm was chosen and optimized to identify and position transmission towers accurately.The thesis goes on to discuss the design and verification of the identification and positioning system of transmission towers on remote sensing maps using deep learning.Introducing the overlap segmentation module and satellite remote sensing image super-resolution module,the thesis aims to enhance the identification effect of transmission towers.For object detection,the thesis constructs a dataset of transmission tower images based on remote sensing,called Dataset of Transmission Tower Images from Remote Sensing Satellites(TIRS).Using the TIRS dataset,the thesis trains the YOLOv5-RSGPS model combined with the geographical information extraction module to locate the transmission towers accurately.Experiments were conducted in the Heitz area of Chicago,where the optimized YOLOv5-RSGPS model had an mAP index of 85.8%and an F1 valueof 83.3%.The average total time taken to identify and position transmission towers was 0.08 seconds.Lastly,this thesis introduces an electronic map processing system and analyzes its feasibility in a New York area field experiment.The map information system was designed based on QGIS and Geoserver,open-source satellite map data was obtained,and image preprocessing was applied to reduce the 26.11 square kilometers recognition area’s identification process to only 16 minutes and eight seconds.The system correctly identified 30 transmission towers but missed two and incorrectly identified one.The F1-Score accuracy index for the area was determined to be 95.24%.The deep learning identification and positioning system of transmission towers on remote sensing maps can accurately and efficiently identify and position overhead transmission towers,fulfilling the accuracy and speed requirements of power grid patrol.