Microimaging-Based AI Classification for Food Quality Assessment in Medical Nutrition

Yue Liu^*

• Guizhou University, Guiyang, China

In the context of advancing microimaging techniques for food quality evaluation, closely aligning with the emerging focus on microstructural visualization and chemical mapping in food science, we recognize the imperative for high-resolution, interpretable imaging solutions, especially in medical nutrition applications. Traditional methods, while effective in basic imaging and chemical assessment, often fall short in resolving the intricate heterogeneity of food matrices, lacking robustness in spectral-structural integration and adaptability to diverse food compositions. Addressing these limitations, we propose a novel framework comprising the Spectral-Structural Fusion Network (SSFNet) and Knowledge-Infused Adaptive Enhancement (KIAE). SSFNet innovatively fuses spectral signatures with spatial microstructures via dual-stream encoding and cross-attention mechanisms, achieving superior material reconstruction accuracy. Simultaneously, KIAE systematically integrates domain-specific priors—compositional constraints, spectral characteristics, and phase behaviors—into the learning process, enhancing model robustness and interpretability. Our methodology not only improves the precision of food microimaging but also strengthens the adaptability of imaging models to a wide range of food types and nutritional profiles, addressing critical challenges in the field emphasized by microimaging community. Extensive experiments demonstrate that our integrated approach significantly surpasses conventional models in spatial resolution, chemical consistency, and noise resilience, offering a powerful tool for food quality control, shelf-life evaluation, and medical nutrition research.