Aiming at the problems appearing at the scene of the unmanned aerial vehicle (UAV) checking transmission lines, such as high requirements for hardware, long running time of the image recognition algorithm and the shortage in the number of training pictures, this paper proposes a new method for the daily inspection of transmission lines based on improved YOLOv3 pruning algorithm. The improved algorithm uses the basic YOLOv3 framework, adding the SPP module, extracting multi-scale features in the same convolutional layer and carrying out pruning treatments (reducing the channel numbers and sliming the framework) on the basis of the original YOLOv3. The improved YOLOv3 pruning algorithm is validated by the pictures of actual site inspection of transmission lines. The test result shows that this method has the advantages of less running time, diversity of input images size and lower hardware requirements for the algorithm. And also the accuracy of fault detection is almost unchanged, with the average iteration time of each step reduced by 0.81 s and overall performance improved by 25%.