A highly efficient and interpretable framework for high-precision lithological identification integrating SERS and...

Area of Science:

• Geochemistry
• Spectroscopy
• Artificial Intelligence

Background:

• Traditional lithological identification methods are slow, destructive, and subjective.
• Accurate rock identification is crucial for geological exploration and engineering.

Purpose of the Study:

• To develop a high-precision lithological classification framework using SERS and AI.
• To evaluate and compare the performance of different AI models for rock classification.
• To provide an interpretable and computationally efficient method for real-time geological monitoring.

Main Methods:

• Synthesized silver nanoparticles (AgNPs) for SERS substrate to enhance Raman signals.
• Utilized Finite-Difference Time-Domain (FDTD) simulations to understand electromagnetic enhancement.
• Trained and evaluated six AI models on a self-constructed SERS dataset of eight rock types.
• Employed SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The Extreme Gradient Boosting (XGBoost) model demonstrated superior performance in lithological classification.
• XGBoost mitigated issues of computational latency and training instability common in deep learning.
• The AI model successfully distinguished between rocks with similar spectra by analyzing lattice vibrations and trace signals.
• SHAP analysis confirmed the model's ability to extract intrinsic physical features.

Conclusions:

• The integrated SERS-AI framework offers a reliable and interpretable approach for lithological identification.
• The XGBoost model provides a computationally efficient solution for real-time geological monitoring.
• This method overcomes limitations of traditional techniques in complex construction environments.