Automatic identification of rock discontinuity sets using modified agglomerative nesting algorithm

Identification of rock discontinuity sets is essential for each studied jointed rock slope, and is also an initial step in many existing methods for rock slope stability analysis. This paper presents a new hierarchical agglomerative clustering method using modified agglomerative nesting (MAGNES) algorithm for automatically partitioning discontinuity sets. It is an orientation-based clustering method, and different linkage criteria (single, complete, and average) are incorporated for merging two closest clusters. The performance of MAGNES is tested using a complicate artificial data set, Shanley and Mahtab’s data set, and a real data set from unmanned aerial vehicle (UAV) survey. In addition, the clustering results of four other well-recognized clustering methods are also chosen as comparisons. It shows that the single linkage criterion is inapposite for partitioning orientations and the complete linkage criterion is not robust. Only MAGNES using average linkage criterion (MAGNES_AVG) shows good performance for detecting discontinuity sets. Generally, the main discrepancies among the clustering results lie mainly in the poles at the boundary of two adjacent joint sets. Considering the real data sets are characterized by “ground truth,” the artificial data set with known classification labels is used to further test which method performs better. The number of misclassification points is adopted as an evaluation index, and MAGNES_AVG performs best in partitioning the poles at the boundary of adjacent joint sets. Another advantage of the proposed algorithm is that it is independent of initial parameters, which is user-friendly.