Cross-Instrument Calibration Transfer in LIBS Using Hybrid Spectral Harmonization and Deep Spectral Migration

Laser-induced breakdown spectroscopy (LIBS) is a powerful analytical technique widely used for rapid elemental analysis and material classification due to its minimal sample preparation and near-real-time detection capability. Despite the increasing availability of LIBS spectral libraries, their effective reuse across different instruments remains challenging because of instrumental variability. Existing calibration transfer methods fail to capture nonlinear spectral discrepancies between instruments. To address these limitations, this study proposes a framework termed Hybrid Spectral Harmonization and Deep Spectral Migration (HSH-DSM) for cross-instrument spectral transfer in LIBS. The proposed approach integrates physically motivated spectral preprocessing with data-driven deep learning to reduce cross-instrument discrepancies. Then, a local intensity harmonization strategy is applied to correct local spectral response differences. Finally, a deep spectral migration model based on a tunable multilayer perceptron (MLP) is trained to learn the nonlinear mapping between a source domain and a target domain. Experimental results demonstrate strong spectral reconstruction performance. The model achieves coefficients of determination (R2) of 0.999 for calibration, 0.990 for validation, and 0.983 for independent test data, indicating accurate cross-instrument spectral translation. To further evaluate the practical effectiveness of the framework, the performance evaluation on unseen target-domain samples yields 97.8% precision, 97.5% recall, and 97.5% F1-score, demonstrating promising classification capability. These results confirm that the proposed HSH-DSM framework effectively mitigates cross-instrument spectral discrepancies while preserving discriminative spectral features. The proposed strategy provides a promising approach toward instrument-independent LIBS spectral libraries and facilitates reliable deployment of LIBS-based analytical models across heterogeneous instruments.