Precise geographical tracing of ginseng using laser-induced breakdown spectroscopy and hyperspectral imaging data fusion in small-sample conditions

Accurate geographical origin authentication of ginseng is pivotal for ensuring medicinal quality and safeguarding consumer rights, yet it remains technically challenging under small-sample conditions. This study proposes a multimodal analytical framework that synergistically integrates laser-induced breakdown spectroscopy (LIBS) and hyperspectral imaging (HSI) to overcome the inherent limitations of single-modality spectroscopy. High-dimensional LIBS and HSI data were first compressed and denoised by principal component analysis (PCA), which retained >90 % variance while reducing the feature dimensionality to 20 principal components. Subsequently, three classical classifiers—support vector machine (SVM), random forest (RF) and k-nearest neighbor (KNN)—were constructed and benchmarked. The PCA–RF model trained on the fused dataset achieved a perfect classification accuracy of 100 %, significantly surpassing the best single-modality models (91.7 % for LIBS and 83.3 % for HSI). The complementary nature of the two modalities was further corroborated by t-SNE and PCA visualizations, which revealed a more compact and linearly separable manifold structure in the fused feature space. Additionally, the key LIBS element emission lines and HSI molecular response regions for differentiation are further determined. These findings elucidate the elemental-to-molecular mechanisms underpinning origin-specific signatures. Overall, the proposed fusion strategy not only fully exploits the complementary strengths of LIBS and HSI but also furnishes an extensible, interpretable and high-precision framework for ginseng provenance authentication, offering robust technical support for quality control in herbal medicine markets.