Traceability of black tea origin by synergistic application of electronic tongue and hyperspectral imaging combined with a Transformer–graph network

Abstract BACKGROUND The quality and commercial value of black tea are significantly influenced by its geographical origin. Traditional traceability methods for black tea are often time‐consuming, complex, and inefficient. This study proposes a novel method for the rapid geographical origin traceability of black tea by integrating an electronic tongue (ET) and hyperspectral imaging (HSI) combined with an improved Transformer–graph fusion network (MSTNet). First, taste and spectral image fingerprints of black tea samples are collected by using ET and HSI systems, respectively. To address the complexity and redundancy of ET signals, a composite exponential weighting strategy is employed to optimize the feature representation, followed by a multi‐scale parallel fusion Transformer (MPFT) to extract temporal features from ET signals. Meanwhile, given the inherent high dimensionality within HSI images, principal component analysis (PCA) is applied to select informative components, after which a spatial‐enhanced Swin Transformer (SEST) is used to capture spatial features from HSI images. Subsequently, a novel graph network is proposed to achieve multi‐source feature fusion and classification. RESULTS The experimental results indicate that the proposed method achieves superior recognition performance, with a classification accuracy of 99.07%. CONCLUSION This study provides a novel method for black tea origin traceability, which offers broad application prospects for the traceability detection of other food industries. © 2026 Society of Chemical Industry.