MC-H-Geo: A Multi-Scale Contextual Hierarchical Framework for Fine-Grained Lithology Classification.

Area of Science:

• Geoscience
• Remote Sensing
• Artificial Intelligence

Background:

• High-resolution lithological mapping is crucial for reservoir characterization and petroleum geology.
• Distinguishing lithologies with subtle petrophysical contrasts from outcrop data presents a significant challenge.

Purpose of the Study:

• To develop an automated framework for lithological classification from terrestrial laser scanning (TLS) point clouds.
• To enhance the accuracy and geological consistency of outcrop interpretation.

Main Methods:

• Proposed MC-H-Geo: a multi-scale contextual hierarchical framework.
• Integrated a multi-scale contextual feature engine, gated expert classifier, and geological post-processing.
• Utilized spectral, geometric, and textural descriptors with cross-scale differentials and stratigraphic continuity enforcement.

Main Results:

• Achieved state-of-the-art performance on the Qianwangjiahe outcrop (OA = 94.3%, Macro F1 = 0.944).
• Outperformed existing methods like PointNet++ (77.1%), SG-RFGeo (74.2%), and XGBoost (61.7%).
• Identified sandstone-vegetation confusion in weathered zones as a limitation of TLS-only data.

Conclusions:

• MC-H-Geo establishes an advanced framework for fine-grained lithological mapping.
• Multi-sensor data fusion is identified as a promising approach for robust, geologically consistent outcrop interpretation.