Integrated Multispectral and Multilevel Data Optimization for Rapid Origin Tracing and Quality Assessment of Salvia Miltiorrhiza

Abstract

The accessment of the herbal medicine quality often involves time-consuming and labor-intensive high-precision instrument testing. This study use Salvia miltiorrhiza as a case study, with 120 samples collected from various regions. Hyperspectral data, Fourier Transform Near-Infrared data, heavy metal data, and active ingredient data were systematically gathered. Using multivariate statistical analysis, multimodal information integration, and deep learning algorithms, a geographical origin tracing model was developed, achieving an accuracy rate of 100%. Additionally, spectral-ingredient prediction models for five heavy metals and four active components were established, allowing the heavy metals and active components of Salvia miltiorrhiza to be rapidly and accurately assessed. This study compared the performance of the hyperspectral and FT-NIR techniques in herbal medicine quality control. The results indicated that hyperspectral demonstrated broader applicability and superior performance in predicting heavy metal content, whereas FT-NIR was more effective in analyzing chemical constituents. This research provides a scientific basis for Salvia miltiorrhiza origin tracing and quality prediction while also confirming the potential of multispectral techniques in the quality accessment.

Keywords: Salvia miltiorrhiza, Heavy metals and harmful elements, Hyperspectral, FT-NIR, deep learning