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

Area of Science:

• Food Science
• Analytical Chemistry
• Artificial Intelligence

Background:

• Hyperspectral imaging (HSI) provides detailed spectral and spatial data for food quality analysis.
• Increasingly complex chemometric, machine learning, and deep learning models in HSI raise interpretability concerns.
• Explainable artificial intelligence (XAI) offers methods to clarify model operations and decisions.

Purpose of the Study:

• To review advancements in HSI for food quality evaluation.
• To explore the role of XAI in enhancing the interpretability of HSI models.
• To discuss challenges and future directions for HSI-XAI frameworks in food systems.

Main Methods:

• Introduction to HSI operational principles and data analysis.
• Discussion of chemometric, machine learning, and deep learning models for HSI.
• Explanation of six prominent XAI methods: SHAP, LIME, Grad-CAM, saliency maps, DeepLIFT, and TCAV.

Main Results:

• XAI methods can illustrate HSI model inputs and outputs, clarify mechanisms, and validate decisions.
• Applications of XAI-enhanced HSI span various food systems, improving transparency.
• Identified challenges include standardization, environmental alignment, and human-in-the-loop integration.

Conclusions:

• Integrating XAI with HSI is crucial for developing interpretable food quality evaluation systems.
• Future work should focus on standardized protocols and human-AI interaction for practical factory-floor applications.
• Reliable HSI-XAI frameworks are needed for actionable and interpretable food quality decisions.