A Scale-Aware Machine Learning Framework for Automated GSI Estimation: From Terrestrial to Planetary Environments

Electronic Supplementary Materials (ESM) for the research paper titled: "A Scale-Aware Machine Learning Framework for Automated GSI Estimation: From Terrestrial to Planetary Environments", submitted to Rock Mechanics and Rock Engineering. This repository contains the dataset, source codes, and numerical results required to reproduce the findings of the study. Contents:

ESM_1_Dataset.zip: Contains 400 rock mass images organized into two domains:

Terrestrial_Images: 200 expert-annotated images from Earth (various lithologies), curated from publicly available geotechnical literature as detailed in Table 1 of the manuscript.

Martian_Images: 200 raw images from NASA's Curiosity and Perseverance rover missions used for zero-shot transfer testing.

ESM_2_SourceCode.zip: MATLAB source codes for the proposed framework. Includes scripts for:

MobileNetV2 feature extraction.

Scale Factor (SF) calculation and integration.

Training the Dynamic Ensemble Selection model (GBM, SVR, KNN).

Martian GSI prediction (zero-shot transfer).

ESM_3_NumericalData.zip: Contains CSV and Excel files:

Ground truth labels (GSI scores) and Scale Factors.

Validation results comparing author-assigned vs. literature labels.

Model prediction outputs for the Martian dataset.