The invention relates to a power transmission line bolt detection method based on a convolutional neural network. The method comprises the following steps: 1) establishing a bolt image library having a label as a sample set; 2) obtaining bolt candidate regions in a bolt image through a sliding window method; 3) carrying out coarse screening on the bolt candidate regions through a SURF algorithm; and 4) serving the coarsely-selected bolt candidate region as input and inputting the coarsely-selected bolt candidate region to the constructed CNN (convolutional neural network) for judgment, judging whether the coarsely-selected bolt candidate region is a correct bolt region according to output vector of the CNN (convolutional neural network), and carrying out bolt missing detection on the bolt region. Compared with the prior art, the method has the advantages of simple and easy-to-implement algorithm flow, fast speed, high recognition rate and good detection effect and the like.