3D Point Cloud Lithology Identification Based on Stratigraphically Constrained Continuous Clustering

To address the challenges of boundary ambiguity and classification accuracy degradation in lithology identification of outcrop point clouds within complex geological settings, this study proposes a Stratigraphy-Guided Continuous Clustering (SGCC) method. Leveraging sedimentological principles of lateral continuity, a dynamic density-threshold hierarchical clustering algorithm is designed to optimize lithological unit boundaries through adjacency-based cluster merging criteria. A Patch-level Feature Aggregation Operator (PFAO) is introduced to construct a multimodal feature space by integrating geometric covariance matrices and spectral distribution entropy. A random forest classifier is then employed for lithology discrimination. Experimental validation on the Qingshuihe Formation outcrop dataset from the Junggar Basin, Xinjiang, demonstrates that the SGCC method achieves an overall accuracy (OA) of 94.64% and a mean intersection over union (MIOU) of 90.87%, outperforming traditional machine learning (SVM, XGBoost) and deep learning methods (PointNet) by 26.22%–68.36%. Notably, SGCC significantly enhances boundary recognition in sandstone-mudstone thin interbeds and conglomerate-sandstone transitional zones. Ablation experiments confirm the efficacy of stratigraphic constraints in suppressing noise and improving computational efficiency, reducing training memory by 83.3% and processing time by 85.7%. By deeply integrating geological principles with computational models, this method provides a high-precision and interpretable technical pathway for intelligent geological exploration.