Abstract: Accurate acquisition of the orientation and spatial distribution of rock mass discontinuities is essential for the stability analysis of high and steep rock slopes. Conventional clustering-based methods, which rely heavily on point cloud normal vectors, often suffer from significant identification errors when abrupt variations exist in the point cloud data. To address this limitation, a discontinuity mapping approach integrating geometric and signal features of point cloud is developed, improving both the accuracy and robustness of the identification result. Furthermore, the k-means++ algorithm is enhanced by refining its decision-making process and optimizing cluster center selection, enabling more effective segmentation and information extraction of rock mass discontinuities. The proposed method is applied to a rock slope in Zhejiang, China. The results demonstrate that it can effectively identify discontinuities from point cloud data, accurately extract their orientations, and perform dominant set classification. Comparative analysis with the DBSCAN algorithm further verifies the effectiveness of the proposed approach. Validation against manual measurements shows that the errors in dip direction and dip angle are within 3° and 4°, respectively, indicating that the method provides accurate and reliable results with strong engineering applicability.

Key words: rock slope, UAV measurement, 3D point cloud, discontinuity identification, point cloud feature, clustering algorithm