A Multilevel Classification Strategy for the Identification of Discontinuities from 3D Point Clouds of Complicated Rock Surfaces

In natural environments, open slopes commonly exhibit complicated rock outcrops, which present significant challenges for the measurement of rock discontinuities. This research employs UAV laser scanning to gather data on rock outcrops within complex topographies and proposes a multilevel classification strategy (MLCS) to identify the discontinuities using slope point clouds that encompass multiple geographic elements. The MLCS applies Random Forest (RF) to extract rock discontinuity point clouds from complex slopes (Level 1) and then conducts a preliminary segmentation of the main direction sets of the discontinuity point clouds using an improved kernel density estimation (IKDE) algorithm (Level 2). Finally, the density-based spatial clustering of applications with noise (DBSCAN) method is used to refine the segmentation of individual discontinuities (Level 3). In this strategy, RF focused on the training features of planar structures is selected to preprocess the slope point cloud. This approach accurately distinguishes discontinuities from other geographic elements, reducing the impact of complex environments. Additionally, two cases were used to validate the IKDE, which entailed consistency with the previous research results (Case 1) and the preservation of the integrity of rough and curved discontinuities (Case 2). Both of these cases demonstrate the strong adaptability of automated IKDE. These findings indicate that the proposed MLCS effectively measures discontinuities in complex rock surfaces in natural environments.