Astragalus membranaceus, a widely recognized traditional Chinese medicinal herb, is extensively employed for its immunomodulatory and health-enhancing properties. The quality and therapeutic efficacy of Astragalus are profoundly influenced by its geographical origin, underscoring the necessity for reliable methods to authenticate its provenance, ensure product integrity, and mitigate adulteration risks. Conventional identification techniques, encompassing morphological, chemical, and DNA-based approaches, are often constrained by their time-intensive, laborious, and costly nature, thereby limiting their applicability in large-scale industrial contexts. Spectroscopic techniques, such as Laser-Induced Breakdown Spectroscopy (LIBS) and Near-Infrared Spectroscopy (NIR), have emerged as rapid, non-destructive, and efficient analytical tools for quality assessment and geographical origin determination. Nevertheless, the inherent complexity of Astragalus, characterized by its diverse elemental and molecular profiles, often renders single-spectral techniques inadequate for comprehensive characterization. Data fusion methodologies, which integrate complementary information from multiple sources, offer a promising avenue to enhance classification accuracy. By leveraging advanced data fusion strategies to combine LIBS and NIR spectral data, the accuracy of geographical origin discrimination for Astragalus membranaceus can be substantially improved.Astragalus samples were collected from five different geographical origins: Gansu, Heilongjiang, Inner Mongolia, Shanxi, and Shaanxi (Fig.1). Complementary elemental and compositional information was obtained using LIBS and NIR techniques (Fig.2). Initially, Support Vector Machine (SVM), Logistic Regression (LR), and Linear Discriminant Analysis (LDA) models were developed based on individual LIBS and NIR spectral data, and LDA was selected as the base model for investigating fusion classification outcomes based on the single-spectral classification results (Tab.1). To improve classification performance, lower-level and mid-level data fusion strategies were employed to integrate LIBS and NIR spectral information. Lower-level data fusion involves the direct concatenation of LIBS and NIR spectral data to form a new lower-level fused spectral dataset for model classification (Fig.4). Mid-level data fusion, on the other hand, extracts the most representative features from LIBS and NIR spectra separately and then concatenates these features to form a mid-level fused spectral dataset for model classification (Fig.4). The model's performance was evaluated using various metrics, including classification accuracy (ACC), macro-precision (M-P), macro-recall (M-R), macro-F1 score (M-F1), and the Area Under the Curve (AUC), to assess the effectiveness of spectral fusion strategies compared to single-spectral approaches.In single-spectrum analysis, the LDA model for LIBS achieved an optimal classification accuracy of 88% on the test set (Tab.1). In comparison, the lower-level fusion LDA model attained an accuracy of 92.00% and an AUC value of 0.9964 on the test set (Tab.3). The most notable enhancement, however, was observed in the mid-level fusion approach, which utilized the Stepwise Projection Algorithm (SPA) for feature selection on both LIBS spectral lines and NIR data. This mid-level fusion LDA model achieved a classification accuracy of 96.00% and an AUC value of 0.9998 on the test set (Tab.3), showing substantial improvements in both precision and reliability. The mid-level fusion approach successfully eliminated redundant data, enabling more efficient and accurate classification. Finally, an importance analysis was conducted on the features in the mid-level fusion (Fig.9), with key features being interpreted. The results indicate that integrating complementary spectral data from LIBS and NIR significantly outperforms single-spectrum analysis in terms of classification accuracy and robustness.The results demonstrate the efficacy of combining LIBS and NIR spectral data through data fusion for the accurate and efficient identification of the geographical origin of Astragalus membranaceus. The mid-level fusion model, which integrates feature selection techniques, provided the highest classification performance, indicating its potential for non-destructive and rapid origin authentication. The findings not only highlight the advantages of spectral fusion in enhancing classification accuracy but also propose a reliable and scalable solution for the quality control and traceability of medicinal herbs in the pharmaceutical industry. The successful application of LIBS-NIR spectral fusion paves the way for more comprehensive analytical approaches in the quality assessment of traditional Chinese medicinal materials.