Intelligent identification of rock mass discontinuities based on the MK‐Means algorithm

Abstract Rock mass discontinuities govern the mechanical properties and engineering stability of rock masses. Conventional clustering algorithms typically require manual specification of the number of clusters or initial parameters, rendering the results sensitive to empirical judgment and initial conditions. To address these limitations, this study proposed an enhanced MK‐Means clustering algorithm for discontinuity‐set classification that can automatically determine the number of clusters based on the nonuniform density of the data. Two typical areas on a high, steep rock slope at the lower reservoir of a pumped‐storage power station were selected to implement the proposed method and evaluate its performance. Subsequently, a region‐growing algorithm was applied to segment individual discontinuities from the classified discontinuity sets, and their orientations were extracted. The results indicate that: (1) The proposed method exhibited better classification performance in the two study areas, with global accuracies of 98.28% and 98.46% and mIoU values of 96.48% and 96.87%, respectively, while the maximum errors in dip direction and dip angle were limited to 5° and 4°, demonstrating its high reliability and applicability for discontinuity identification; (2) the effectiveness and rationality of the automatically determined number of clusters were validated through cluster validity indices; (3) visual comparisons with traditional clustering methods using error maps demonstrated that the proposed method yielded more accurate and reasonable discontinuity‐set classification results, further confirming its high applicability.