Feature fusion-enhanced t-SNE image atlas for geophysical features discovery

Area of Science:

• Geophysics
• Data Science
• Geological Exploration

Background:

• Traditional geophysical analysis struggles with complex geology and integrating diverse data types.
• Existing methods often overlook patterns in related datasets, limiting comprehensive geological understanding.

Purpose of the Study:

• To develop a data-driven approach for autonomous discovery of geophysical features from integrated gridded datasets.
• To overcome limitations of traditional tools in handling geological complexity and multi-dataset analysis.

Main Methods:

• Utilized Haralick texture descriptors to encode geophysical data patches into a unified high-dimensional space.
• Applied t-distributed stochastic neighbor embedding (t-SNE) for nonlinear projection into a 2D interactive 't-SNE Atlas'.
• Integrated magnetic and gravity data from the Yilgarn Craton, Western Australia.

Main Results:

• Developed an interactive 't-SNE Atlas' for intuitive navigation and exploration of complex geophysical relationships.
• Successfully revealed subtle geophysical patterns and facilitated the discovery of new geological insights.
• Demonstrated the atlas's utility in identifying geological settings associated with mineral prospectivity.

Conclusions:

• The proposed methodology offers a powerful, adaptable tool for geoscientific exploration by integrating diverse gridded datasets.
• The 't-SNE Atlas' enhances the discovery of geological phenomena and aids in pre-planning exploration activities.
• This approach has broad applicability and can accelerate the identification of promising exploration sites.