Abstract: With the rapid development of the food industry, technologies based on spectral analysis, machine vision, electronic noses and electronic tongues have been widely used in food inspection, effectively improving both the detection efficiency and objectivity. However, food systems exhibit complex and diverse compositions, and single-modal detection technology struggles to meet the requirements for the accurate detection of complex food matrices owing to limitations in information dimensionality. In this context, multimodal data fusion technology has gradually become an important research direction in the field of food inspection. This technology integrates multi-source heterogeneous data such as spectra, images, sensor signals, and chromatographic data to exploit the complementarity and correlation between data sets, thus enhancing the accuracy and intelligence level of food inspection. This paper focuses on research progress on multimodal data fusion technology in the field of food inspection, introduces its fusion levels and modeling methods, and illustrate the current status of the application of this technology in the non-destructive testing of fruits and vegetables, beverage authentication, safety detection of meat and aquatic products, and origin traceability of agricultural products. Moreover, it summarizes the problems existing in current research. In the future, it will be necessary to systematically define applicable conditions for different modality combination and fusion levels and construct a reusable fusion selection framework, thereby providing technical support for the continuous monitoring and traceability management of food quality and safety.

Key words: multimodal data fusion technology; food detection; multi-source heterogeneous data; non-destructive testing; geographical origin traceability