Patrol Operation Center of Guangdong Power Grid Co., Ltd，
As an important part of overhead transmission lines, it is of great significance to find out the wire defects in the process of inspection to avoid the occurrence of serious power accidents.In the process of UAV(unmanned?aerial?vehicle) patrol, the camera collects images of wire defect that have characteristic of high pixel. In order to identify the defects, it needs to process a large amount of data in the computer because its high pixel characteristic. In this paper, the onboard computer of UAV carries out real-time processing of high-resolution wire image collected by HD camera to identify wire defects.A Patch-based method is adopted to firstly conduct down-sampling processing of the wire image and then use u-net as the attention net to obtain the segmentation area of the low-pixel wire and grid it, and then cut out a number of line attention patches. Then, map the line attention Patch to the original high-pixel image and cut out a number of high-resolution line attention patches.The batch input takes mobilev2 as the backbone yolov3 network for wire defect identification, and obtains the original image coordinates of the bounding box of the defect identification target in the relative position of the original image according to the input high-precision focus area, and marks the bounding box of the identified target in the original image.The experimental results show that the method proposed in this paper can realize the defect recognition of high frame rate on the high-resolution wire image collected by camera in the process of UAV line patrol, which provides a new idea for the intelligent patrol of UAV.