Research on image recognition of UAV distribution line inspection based on deep learning

Transmission line is an important part of the power grid, and its running state is related to the safe operation of the power grid. Using deep learning technology to identify defects in photographs taken can not only improve the efficiency of defect discovery and reduce personnel costs, but also has great prospects for deep application of deep learning technology in all aspects of defect identification of power grid transmission lines. Firstly, the method of adding spatial pyramid pooling (SPP) module under the original basic framework of YOLOv3 is proposed. Secondly, a pruning method for channel reduction and frame slimming of YOLOv3 with SPP module is proposed to reduce the calculation time. Finally, the improved algorithm is verified by using a small number of transmission line defect image data sets. The test results show that the improved model has less dependence on hardware, the detection accuracy is almost unchanged, and the average iteration time per step is reduced by 0.76 s.