Authenticity identification of Pinellia ternata based on hyperspectral imaging combined with spectral-texture multimodal data fusion strategy

Pinellia ternata has high commercial and ecological value. As a traditional Chinese medicine (TCM), the authenticity identification of Pinellia ternata is critical to maintaining the quality and safety of the market. Traditional chemical detection methods are highly destructive and operationally cumbersome, struggling to meet industrial-scale rapid screening demands. This study developed a hyperspectral imaging-based spectral-texture multimodal fusion strategy and combined with machine learning algorithms to achieve efficient non-destructive discrimination of TCM, taking Pinellia ternata (PT), Arisaema heterophyllum (AH), Typhonium flagelliforme (TF), and Pinellia pedatisecta (PP) as examples. Hyperspectral images from 1140 samples were acquired in the 898–1751 nm spectral range. Critic weighting method objectively identified key characteristic wavelengths, while 11 textural features were extracted using gray-level run-length matrix analysis. A variety of feature selection strategies were employed to refine texture dimensions. By constructing dual-stage feature screening, important wavelengths and texture information were evaluated and captured. Based on the above, a spectral-textural multimodal dataset was constructed. Models such as Data-Driven Soft Independent Modeling of Class Analogy (DD-SIMCA) algorithm enabled authenticity verification, while the Linear Discriminant Analysis (LDA) were employed for TCM sorting. Results showed that DD-SIMCA model implementing spectral-image data fusion strategy demonstrated superior efficiency (99.87 %) over single-modality analytical approaches (92.47 %). The classification accuracy rate of LDA for PT and its counterfeits reached 99.17 %. Therefore, the hyperspectral imaging-based spectral-texture multimodal fusion strategy and intelligent identification framework overcome the limitations of single-modality analysis, and offered technical foundations for intelligent classification of raw material in manufacturing of TCM. • Authenticity identification of Pinellia ternata using hyperspectral imaging. • A multimodal data fusion strategy based on spectrum-texture information is raised. • Building Critic-NAD feature screening strategy to achieve cooperative optimization. • The DD-SIMCA model can effectively identify the authenticity of Pinellia ternata . • The strategy increased identification efficiency to 99.87 %, an increase of 7.40 %.