Table 2.

Summary of key components in the proposed workflow.

Component	Function in the workflow	Specific advantage	Challenge addressed
Paired PPL & XPL images	Use of raw dual-channel optical inputs	Captures complementary optical and textural features	Overcomes the loss of information typical in single-mode or feature-engineered approaches
YOLO v8	Automated Region of Interest (ROI) detection	Enables efficient, consistent cropping of relevant image areas	Removes subjective bias and reduces manual preprocessing effort
Deep convolutional vision models	Robust feature extraction	High capability of extracting features from images and preserving spatial correlation	Suitable for processing high-resolution petrographic images and extract relevant features
MLP	Final classification layer	Simple, effective method to combine learned features into predicted classes	Supports flexible adaptation to different classification schemes
Optuna	Automated hyperparameter tuning	Ensures optimal model configuration without extensive manual trial	Improves reproducibility and enhances model generalization