Intelligent Identification of Rock Mass Discontinuity Based on a Complexity-Penalized Fuzzy C-Means Algorithm

Natural rock mass discontinuities form mechanically weak zones whose mechanical properties critically influence the stability and safety of engineered slopes. However, in practical engineering scenarios, large numbers of discontinuities cannot be manually surveyed and documented. To address this limitation, this study proposes an intelligent rock mass discontinuity identification method based on a complexity-penalized fuzzy C-means (FCM) algorithm, which automatically retains or eliminates clusters to ensure clustering accuracy. The proposed method is validated using typical discontinuities of a steep rock slope in the lower reservoir area of a pumped-storage power station. After discontinuity set classification, individual discontinuities are further extracted using an improved planar region-growing algorithm. The results demonstrate that (1) the proposed method overcomes the limitation of conventional FCM algorithms that require a predefined number of clusters, optimizes the cluster selection process, reduces computational cost, and significantly improves clustering accuracy; (2) the improved planar region-growing algorithm directly integrates planar geometric characteristics with candidate growth points, enabling the identification of small and fragmented discontinuities embedded within large planar surfaces; and (3) in the case study, the proposed method achieves a global accuracy of 92.15 percent and a mean intersection-over-union of 88.98 percent, indicating reliable clustering performance.