Abstract: Based on the detection and recognition technology of YOLO v3 deep convolution neural network, an automatic recognition method of digital borehole image fractures is proposed. Firstly, the target detection principle of new version YOLO v3 is described in detail, then the borehole image of coal mine is selected to make data sets on VOC 2007, and the network structure of Darknet-53 is used for data training. The experimental results show that the detection method of borehole image fractures based on YOLO v3 can identify the feature information quickly and accurately, which provides a new technical support for the visual recognition of the surrounding rock fractures.