Rapid discrimination of medicinal and industrial mirabilite based on intelligent sensing, near-infrared spectroscopy, and machine learning.

Mirabilite is a widely used mineral medicine with notable purgative and anti-inflammatory activities. However, medicinal mirabilite (MM) and industrial mirabilite (IM) exhibit highly similar appearances, making conventional identification methods unreliable and potentially compromising clinical safety. In this study, a multi-dimensional parallel characterization approach was established for the rapid discrimination and quality evaluation of mirabilite using electronic sensing technologies and near-infrared (NIR) spectroscopy. Each analytical modality was applied independently to characterize distinct physicochemical attributes of the samples without feature-level or decision-level data fusion. Specifically, an electronic eye (E-eye) was used to evaluate visual characteristics, an electronic nose (E-nose) was employed to characterize volatile odor profiles, and NIR spectroscopy was applied to obtain chemical information associated with crystallization water. The E-eye provided limited discriminative capability, whereas the E-nose effectively captured characteristic odor features of MM. For NIR analysis, spectra preprocessed using Savitzky-Golay smoothing with first derivative (SG + 1D) were modeled using partial least squares discriminant analysis (PLS-DA), support vector machine (SVM), and random forest (RF). Among these models, SVM achieved the best classification performance, with accuracy, precision, recall, and F1-score all reaching 100% in both the training and test sets. Furthermore, an eXtreme Gradient Boosting (XGBoost) regression model accurately predicted crystallization water content (R2 = 0.9848, RPD > 3), and SHAP analysis identified the OH combination band around 1900 nm as the most influential spectral region. The proposed approach provides accurate and interpretable discrimination of MM and IM and offers a reliable analytical tool for quality evaluation and safe clinical application.