Explainable AI for hyperspectral imaging in food quality decision support: interpretability, reliability and future directions

Reliable food quality evaluation requires analytical systems that capture both chemical composition and spatial variability while supporting interpretable decisions. Hyperspectral imaging (HSI) has emerged as a technique that provides detailed spectral and spatial information about samples. However, the increasing use of chemometric, machine learning, and deep learning models raises concerns about interpretability. Explainable artificial intelligence (XAI) offers a solution by illustrating inputs and outputs, clarifying model mechanisms, and validating decisions. This review summarizes recent advances in HSI-based food quality evaluation and the role of XAI in improving interpretability. It introduces the operational foundations of HSI, followed by data analysis procedures and representative algorithms and models. Key concepts and categories of XAI are discussed, and six prominent methods are explained, including Shapley Additive exPlanations (SHAP), Model-agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (Grad-CAM), saliency maps, Deep Learning Important FeaTures (DeepLIFT), and Testing with Concept Activation Vectors (TCAV). Applications of XAI-enhanced HSI across food systems are discussed. Challenges are analyzed from food quality, HSI, and XAI perspectives. Future progress will require standardized assessment protocols, rigorous environmental alignment, and human-in-the-loop interfaces to bridge the gap among high-dimensional data, complex models, and actionable factory-floor inspection, establishing reliable HSI-XAI frameworks for interpretable food quality decisions.