Multiple Optical Sensor Fusion for Mineral Mapping of Core Samples

Area of Science:

• Geoscience
• Remote Sensing
• Data Science

Background:

• Geological object analysis is complex due to compositional variability and unclear class boundaries.
• Mineralogical mapping requires specialized processing schemes for diverse geological data.
• Multi-sensor approaches are crucial for overcoming limitations of single optical sensing technologies.

Purpose of the Study:

• To devise an adapted multi-optical sensor fusion (MOSFus) workflow for geological data analysis.
• To account for geological characteristics in a comprehensive processing chain from data acquisition to mineralogical mapping.
• To improve classification accuracy for complex geological objects using fused multi-sensor data.

Main Methods:

• Spatial feature extraction using morphological profiles on high-resolution RGB data.
• Noise reduction for hyperspectral data with mixed sparse and Gaussian contamination.
• Dimensionality reduction via sparse and smooth low-rank analysis.
• Support Vector Machine (SVM) classification for robust handling of unbalanced and sparse training sets.

Main Results:

• The proposed MOSFus workflow effectively fuses heterogeneous data at variable resolutions, scales, and spectral ranges.
• Feature extraction and subsequent processing significantly improve classification performance.
• The approach demonstrates superiority over common strategies in mineralogical mapping tasks.
• Evaluation on two distinct multi-optical sensor datasets confirms the method's effectiveness.

Conclusions:

• The developed MOSFus workflow provides an effective solution for mineralogical mapping of complex geological formations.
• Integrating spatial and spectral information through advanced processing enhances geological data analysis.
• The approach offers a robust and superior alternative to conventional methods for multi-sensor geological data fusion.