Research On The System Based On Embedded Platform And Dual-spectral Imaging Which Can Be Used For Automatic Power Inspection

With the rapid development of our country’s infrastructure,power grids have spread across most areas of the country.The insulators and other components on the high-voltage transmission line are exposed to the outside all year round and are subjected to various natural erosions,often causing problems such as damage and aging,and then causing unnecessary losses to the power system.Moreover,due to the vast land area of our country,many transmission lines are erected in mountains,swamps and other places with very complicated terrain,and are often located at an altitude of tens of meters to hundreds of meters.These factors have put forward a considerable test for power inspections.The traditional manual power inspection method has gradually failed to meet the current power inspection requirements because of its extremely low inspection efficiency,high labor cost and extremely high safety hazards.Since drones can quickly fly over various complex terrains,and can fully realize remote operations of inspectors,the electric inspection method of drones is far higher in work efficiency,labor costs and safety than traditional manual power inspection method.In this regard,this paper takes the principle of abnormal corona discharge of transmission lines as the theoretical basis,and uses HiSilicon’s embedded processor as a platform to design and develop a powerline automatic inspection system based on embedded system and dual-spectral imaging.(1)In view of the abnormal corona discharge of insulators,this paper proposes a method for detecting the position of the corona discharge in the ultraviolet image of the solar blind area.Image processing algorithms such as binarization segmentation,morphological filtering,connected components labeling,etc.are used to suppress the discrete spots of the solar blind ultraviolet light spot image,and the corona discharge situation judgment process is proposed to further eliminate the weak discharge situation.The approximate position of the corona discharge is finally calculated and marked on the image of the visible light channel.Moreover,this article uses the image processing method to directly fuse the image of the visible light and the solar blind zone ultraviolet light,so that the discharge situation is more intuitively reflected on the image of the visible light channel.Finally,experiments have proved that the corona discharge position detection function and the dual-spectral image fusion function have excellent real-time and robustness.The procession time of the corona discharge position detection per frame is about 1.5ms when there is no target,and between 16.5 ms and 18ms when targets are detected.The calculation time of each frame of image fusion is between 4.0ms and 4.9ms.(2)Aiming at the problem of target recognition in power inspection,this paper proposes a target recognition method for insulators and transmission towers.The paper uses the YOLOv3 network model to train aerial pictures of ceramic insulators and power towers,and then successfully transplants them to the HiSilicon embedded platform through model conversion.Due to the insufficiency of processing capabilities of the embedded platform,the target recognition function is optimized for the embedded platform,and the video output is 25 frames per second and the target recognition frame is refreshed at 10 frames per second.Finally,it is proved by experiments that the accuracy rate of the transmission tower identification of this function is 90.27%,and the accuracy rate of the insulator identification is 93.44%.The calculation speed is 12 frames per second.(3)In response to the current demand in the field of power inspection,this paper designs a dual-mode system basic framework.The system uses a dual-spectrum image acquisition system to separately collect visible light and solar-blind ultraviolet light.The video signal is processed and output in two different ways in the embedded system.The first method is the MIPI LCD screen visual output,which is used in handheld mode.The second method is H.264 encoding and then using Ethernet for output,which is used for conmmunication with drones when it is loaded with UAV in the future.