A systematic review of multimodal fusion technologies for food quality and safety assessment: recent advances and future trends

Background

Food safety and public health are increasingly threatened by the frequent occurrence of food adulteration, microbial contamination, and excessive pesticide residues. These challenges have become increasingly severe and pervasive. In recent years, the advancement of artificial intelligence has promoted the development of unimodal approaches. However, the performance of these methods is limited due to the reliance on single-dimensional information. The growing emergence of multimodal fusion technology is expected to substantially improve the detection of complex food samples.

Scope and approach

This paper presents a systematic review of multimodal fusion methods recently employed in the detection of food quality and safety. It first reviews the characteristics and applications of commonly used detection modalities, followed by a detailed analysis of the implementation mechanisms of fusion strategies at three distinct levels: low-level, mid-level, and high-level. Additionally, it summarizes the status of these strategies across typical tasks such as food adulteration identification, origin traceability, and flavor modeling. Finally, the paper highlights current technical challenges and proposes potential directions for future research.

Key findings and conclusions

Multimodal fusion methods hold significant potential to enhance the accuracy and stability of food detection models. Compared to traditional unimodal approaches, multimodal strategies offer clear advantages in detecting food quality and safety. These benefits come from their ability to combine diverse types of information from multiple sources in a more effective way. However, current multimodal fusion methods still encounter certain challenges in practical applications. With ongoing advancements in artificial intelligence technology, multimodal fusion methods are expected to become more widely adopted in food safety detection. This broader application will provide robust support for food safety supervision.