Digital borehole camera technology can accurately obtain information about the characteristics of the structural surface of the rock in the borehole. To address the shortcomings of the existing digital borehole image analysis, which is labor-intensive, subjective, and computationally intensive, a new analysis scheme is proposed to automate the recognition of borehole interior images captured by digital borehole camera technology. Firstly, the digital borehole images are pre-processed and the edge features the images are extracted using a pre-trained DexiNed network; secondly, the Epremoval method is proposed to deal with the edge point noise and extract the region of interest; finally, according to the Taylor expansion of the sine curve, this method performs polynomial fitting on the characterization data in the image. The parameters of the rock structure surface are obtained by calculating, spatial transformation and mathematical transformation of the obtained curves. The digital borehole image of a tunnel project is used as an example to apply the above algorithm. The obtained results are compared with the results of manual assisted interpretation, and the comparison results show that the recognition is better.