Device-Robust Spectral Grading and Origin Detection from UV-Vis-NIR Images: Towards Practical Gemstone Quality Assessment

Mahule Roy, Srikanth Thudumu, Jason Fisher, Rajesh Vasa, Kon Mouzakis

Abstract: Assessing gemstone quality and provenance traditionally requires controlled lighting, standardized devices, and subjective expert analysis. We propose a robust, device-robust framework for automated gemstone assessment that decouples visual quality grading from origin verification. First, for quality assessment, we introduce a colorimetric grading pipeline using multi-modal UV-Visible-NIR imaging and CIELAB analysis. To ensure cross-device consistency, we employ a perceptual color difference metric ($\Delta E$) with calibrated thresholds. We address the challenge of small-data regimes in gemology via synthetic color perturbations that expand minority classes while preserving perceptual realism. A spectral fusion module further integrates RGB, UV fluorescence, and NIR reflectance imagery, guided by segmentation masks to enhance grading robustness. Second, for geographic origin verification, we introduce a spectroscopy-assisted, hybrid pipeline that matches UV-Vis-NIR spectral signatures against reference templates using interpretable signal-processing techniques. This component compares the spectral signature of a specimen against a library of known-origin templates using Normalized Cross-Correlation and Dynamic Time Warping (DTW) to account for non-linear spectral shifts. We evaluate the framework using a ``Leave-One-Stone-Out'' cross-validation protocol to prevent data leakage. Experimental results demonstrate that our imaging approach maintains grading accuracy across varying camera hardware (retaining $>90\%$ accuracy), while the uncertainty-aware rejection mechanism flags ambiguous cases. The proposed framework demonstrates a proof-of-concept, interpretable approach for gemstone assessment, highlighting the feasibility of device-robust color grading and hybrid spectral origin screening under controlled experimental conditions.