Explainable Artificial Intelligence Techniques for Interpretation of Food Models: a Review

AI is being used to ensure food safety and quality, but consumers and inspectors need to understand how these systems make decisions. The paper notes that AI models using spectral imaging can detect contaminants or assess freshness, but their "opaque decision-making process hinders adoption"—meaning people are hesitant to trust AI recommendations they cannot verify.

Explainable AI (XAI) techniques can show exactly which factors influence food quality decisions. Specific methods like SHAP and Grad-CAM can identify which spectral wavelengths or image regions a model relies on when assessing food, allowing quality control inspectors to verify whether the AI's reasoning makes sense before acting on its recommendations.

XAI remains underutilized in food engineering despite its potential benefits. The paper identifies this as a significant gap, suggesting that the food industry has not yet widely adopted transparency tools that could build confidence in AI-driven quality control systems and meet "stringent food quality standards."

A structured framework is needed to help food engineers choose the right explanation methods for their specific problems. The paper presents a taxonomy organizing XAI techniques by data types and explanation methods, recognizing that different food quality applications may require different approaches to make AI decisions interpretable.