3D point cloud lithology identification based on stratigraphically constrained continuous clustering.

Area of Science:

• Geological Sciences
• Computational Geology
• Remote Sensing

Background:

• Automated lithology identification from 3D laser scanning data faces challenges in areas with indistinct boundaries and complex stratigraphy.
• Existing methods struggle with accuracy in transition zones and complex geological formations.

Purpose of the Study:

• To develop and validate a novel framework, Stratigraphically Constrained Continuous Clustering (SCCC), for high-precision lithology identification in geological outcrops.
• To improve classification accuracy and boundary delineation in challenging geological settings.

Main Methods:

• The SCCC framework integrates sedimentological principles of lateral continuity with a dynamic density-threshold hierarchical clustering algorithm.
• A patch-level feature aggregation module constructs a multimodal feature space using geometric covariance matrices and spectral distribution entropy.
• Lithology discrimination is performed using a Random Forest classifier.

Main Results:

• SCCC achieved 94.64% overall accuracy, 94.58% F1-score, and 90.87% mean intersection over union on the Qingshuihe Formation outcrop dataset.
• The method significantly outperformed traditional machine learning and deep learning approaches by 26.22-68.36%.
• Stratigraphic constraints improved computational efficiency, reducing memory usage by 83.3% and processing time by 85.7%.

Conclusions:

• The SCCC framework offers a high-precision and interpretable pathway for intelligent geological exploration.
• It demonstrates superior performance in identifying lithologies and delineating boundaries, particularly in transition zones and thin interbeds.
• The integration of geological principles with computational models enhances the robustness and efficiency of outcrop analysis.