As an important component of overhead transmission lines, transmission wires may cause major power accidents if they are found to be defective during the inspection process. The identification of wire defects in high-pixel UAV inspection images requires a large amount of calculation and the target area of defects is small. If real-time high-precision identification is realized in the UAV inspection process, the inspection efficiency can be greatly accelerated. This paper proposes a method to identify wire defects by using the on-board computer of UAV to process high-resolution wire images collected by high-definition cameras in real time. The method firstly uses semantic information patching to sub-sample the wire image, and then uses the segmentation network to obtain the low-pixel wire segmentation areas and mesh them. Multiple wire areas are cropped out and mapped back to the high-pixel original image at a down-sampling ratio. Then a number of high-resolution wire areas are cut out and enter the yolov3 network in batch for wire defect recognition. Finally the defect recognition target area is produced according to the input high-precision attention area in the relative position of the original image. The experimental results show that the identification method proposed in this paper can realize the real-time identification of high frame rate defects in the high-resolution wire images collected by the camera in the process of UAV patrolling, which provides a new idea for the intelligent patrolling of UAV.