Research On UAV Vision-based Distribution Lines Inspection System

The operation and maintenance of the distribution network is related to the economic benefits of the electric power sector and the people’s life with electricity.The regular inspection of distribution lines is an important link to maintain the normal operation of the distribution network.The distribution lines have a wide coverage and long overhead lines,and the most commonly used manual inspection methods have safety hazards and low efficiency.The proposed UAV vision-based distribution power line inspection system uses image processing and deep learning methods to detect distribution lines information and distribution towers information in UAV aerial images.Calculate the relative angle between the center of the UAV image and the distribution lines as deviation data by fusing the distribution lines information and distribution towers information in the image.UAV vision-based distribution lines detection is an important link used for UAV distribution lines tracing.By sending the deviation data calculated in this study to the UAV flight controller,UAV distribution lines tracing can be realized.The main research work is as follows.(1)The scheme of distribution lines inspection system is proposed.The complete UAV and cell phone terminal in the system are selected,and the cell phone terminal software is developed.(2)Realized distribution lines inspection by using image processing algorithms.Image pre-processing of the UAV aerial images to reduce the amount of subsequent processing calculations and interference in the background.The LSD algorithm is used to detect the line segments in the pre-processed images,and the LSD detection results are combined and stitched to obtain the inspection results of the distribution lines in the UAV aerial images.Finally,the distribution line trajectory is obtained by fitting the inspection distribution lines,and the distribution line trajectory and the distribution towers inspection results are fused to calculate the deviation data used for the UAV flight heading control.(3)Applying Mobile Net V2 and Deep Labv3+ algorithms to distribution lines inspection.Building a semantic segmentation dataset for distribution lines,and the network is improved and optimized for the distribution lines inspection task.The improved network improves the Intersection-over-Union index by 2.85% on the dataset.The YOLOV5 s target detection network is applied to distribution towers inspection in UAV aerial images of distribution lines images.Building a distribution towers target detection dataset and the training dataset is data augmented.The network detection average precision index reaches 99.5% on the dataset.(4)Testing the algorithm with manually labeled videos and use the UAV to flight test the distribution lines in the actual environment.The experimental results show that the system can be used to achieve UAV distribution lines tracing.