Identification of rock mass discontinuity from 3D point clouds using improved fuzzy C-means and convolutional neural network

Accurately obtaining rock mass discontinuity information holds particular significance for slope stability analysis and rock mass classification. Currently, non-contact measurement methods have increasingly become a supplementary means to traditional techniques, especially in hazardous and inaccessible areas. This study introduces an innovative semi-automatic method to identify discontinuities from point clouds. A modified convolutional neural network, AlexNet, was established to identify discontinuity sets. The network consists of five convolutional layers and three fully connected layers, utilizing 1 × 3 normal vectors computed by K-nearest neighbor and principal component analysis as input and generating an output value “i” that represents the identified discontinuity set associated with the “i” category. Learning samples for network training were randomly selected from point clouds and automatically categorized using the improved fuzzy C-means (FCM) based on particle swarm optimization (PSO). The orientations of individual discontinuities, identified from the discontinuity set using hierarchical density–based spatial clustering of applications with noise, were calculated. Two outcrop cases were employed to validate the efficacy of the proposed method, and parameter analysis was conducted to determine optimal parameters. The results demonstrated the reliability of the method and highlighted improvements in automation and computational efficiency.