Rock discontinuities are crucial to the stability of rock mass. Under complex high-steep slope conditions, such as the interference of vegetation, a large number of discontinuity sets, and a high degree of weathering, the characterization of rock discontinuities is usually challenging. To this end, this paper proposed an approach for rock discontinuity identification and characterization based on UAV photogrammetry and artificial neural network (ANN) algorithm. UAV photogrammetry was used to obtain 3D point clouds of the study area. The vectors of multi-dimensional features including point cloud orientation features (Normal vectors), geometric features (Anisotropy, Planarity, Roughness, etc.), and optical features (RGBVI), were obtained by calculation. Then, by constructing an ANN model and using the multi-dimensional feature vectors as the network input, the multivariate classification task of rock discontinuities was realized. The ANN algorithm can simultaneously identify and classify all discontinuities of the rock mass and non-discontinuities, as well as each rock discontinuity set. On this basis, the extraction of the individual discontinuities was achieved by the Fast-marching approach. Two case studies were utilized to illustrate the methodology. The results show that this approach has high accuracy and high computational efficiency. The main advantage of the proposed approach is that the ANN can handle complex discontinuity extraction tasks without a complicated pre-processing process, making it highly applicable.