Objective To achieve rapid traceability detection of soybeans from different geographical origins.Methods A novel method for rapid soybean origin traceability is proposed based on the fusion of information from electronic tongue and electronic eye. The electronic tongue and electronic eye are used to collect taste fingerprint and visual image information, respectively, from soybeans of different origins. To address the differences between the temporal sequence features of electronic tongue signals and the spatial image features of electronic eye data, Informer and Swin Transformer are respectively employed to extract key features from the two types of data. A dual-modal feature interaction module is designed to perform deep interaction between the two types of features, thereby obtaining multi-scale inter-layer mutual information. A cross-attention fusion module is then applied to integrate the inter-layer mutual information, enabling efficient cross-modal information fusion and classification recognition.Results The proposed model has good discriminative ability for soybeans from different origins, achieving an accuracy of 98.33%, precision of 98.34%, recall of 98.33%, and F₁-Score of 0.983 3.Conclusion This method provides an efficient and reliable new approach for the rapid traceability detection of soybean geographical origin.