Handheld Raman spectroscopy offers a rapid, nondestructive method for on-site analysis. However, Raman spectra collected from different instruments often exhibit significant nonlinear discrepancies due to variations in optical configurations, detector responses, and assembly tolerances. These discrepancies hinder data comparability and limit the widespread integration of handheld Raman systems. In this work, we present a standard-material-based model transfer approach to achieve accurate spectral alignment between handheld and benchtop microscopic Raman instruments. The method integrates wavenumber calibration, multiscale convolution, and an attention-guided segmented transfer strategy to establish a high-fidelity mapping from handheld spectra to microscopic Raman reference profiles. This approach effectively corrects nonlinear wavenumber shifts while preserving key spectral features, including peak positions, relative intensities, and overall spectral shape. Experimental results demonstrate that the proposed framework significantly reduces peak position root-mean-square error (RMSE) across diverse analytes and enhances standard spectral similarity metrics, such as Spectral Angle Mapper (SAM), Dynamic Time Warping (DTW), and Maximum Mean Discrepancy (MMD). Additionally, the transferred spectra exhibit improved performance in downstream classification tasks, confirming enhanced cross-instrument reliability. This work provides a robust solution for achieving spectral consistency across Raman platforms, supporting interoperable analysis across different measurement environments and laying the foundation for multi-instrument spectral harmonization.
Li et al. (Wed,) studied this question.