# Dairy quality AI needs an explanation as well as a prediction > A review connects spectroscopy, imaging and electronic sensing with the practical problem of deciding when a model’s answer can be trusted. URL: https://foodradar.org/dairy-quality-ai-sensors-explainability Language: en Author: FoodRadar Editorial Published: 2026-10-07T11:28:09.685Z Updated: 2026-10-07T11:28:09.685Z Source publication date: 2026-08-20 Category: Technology Türkçe: https://foodradar.org/tr/dairy-quality-ai-sensors-explainability ## Matching the sensor to the dairy product Artificial intelligence can help interpret dairy measurements quickly, but the useful signal differs between raw milk, powder, cheese and yogurt. An August review maps sensing technologies to those product-specific tasks and examines how explainable models could support quality decisions. Its argument goes beyond achieving a high classification score: operators need to understand what the system has measured and how reliable its conclusion is. The paper synthesises existing studies rather than introducing a newly validated inspection platform. It covers composition, adulteration, freshness, microbial screening and visible defects. Those applications should remain distinct. A system trained to recognise an abnormal spectrum in milk is not automatically validated to identify the contaminant, inspect cheese structure or determine a yogurt’s remaining shelf life. ## Combining measurements can fill information gaps Spectroscopy can reveal chemical information but may provide limited spatial detail. A camera captures appearance and surface defects, while electronic noses and tongues produce patterns associated with volatile or taste-related characteristics. The review describes combining these complementary signals to build a more complete assessment than a single instrument can provide. More data do not remove the need for a sound reference. Composition labels may require laboratory chemical analysis, microbial labels require culture and identification, and image defects require careful annotation. Rare contamination or adulteration events are particularly difficult to represent in training datasets. A model can consequently fit the available examples well while remaining poorly prepared for an unfamiliar abnormal sample. ## Explanation must be checked against dairy chemistry The review distinguishes explanations of influential inputs, model decisions and the confidence of outputs. A quality team may need to know which spectral regions or sensor signals drove a result, as well as whether the prediction is sufficiently certain to support an action. Confidence intervals or risk categories can be more informative than an unexplained pass-or-fail label. However, a feature-attribution chart does not prove a physical mechanism. The authors note that explanation quality lacks consistent evaluation and should be checked for faithfulness to the model, stability and relevance to dairy chemistry. A plausible-looking explanation can still mislead if it reflects a spurious relationship in the training data rather than the property being tested. ## Reliability has to survive a change of conditions Data scarcity, changing product characteristics and transfer between instruments remain barriers to wider use. The review points toward standardised datasets, multimodal modelling and better evaluation of explainability, while presenting these as development priorities. It does not establish that a single general model already performs every dairy inspection task reliably. For manufacturers, the resulting selection criteria are practical: define the product and defect, establish the reference method, test beyond the development dataset and specify how uncertain results are handled. AI-assisted sensing can support faster quality assessment, but its operational value depends on that complete validation chain. The paper provides a framework for asking those questions before a laboratory result becomes a factory control decision. ## Sources and rights - [Artificial Intelligence-Driven Dairy Quality Assessment: From Advanced Sensing Technologies to Explainable Intelligence.](https://europepmc.org/articles/PMC13512615?utm_source=foodradar.org&utm_medium=referral&utm_campaign=editorial&utm_content=dairy-quality-ai-sensors-explainability-en). Multidisciplinary Digital Publishing Institute (MDPI). Song X, Liang Z, Cheng C, Wu R, Dong G, Cui X, Ding H. Artificial Intelligence-Driven Dairy Quality Assessment: From Advanced Sensing Technologies to Explainable Intelligence. Foods (Basel, Switzerland). 2026. DOI: 10.3390/foods15162925. © 2026 by the authors. [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/?utm_source=foodradar.org&utm_medium=referral&utm_campaign=editorial&utm_content=dairy-quality-ai-sensors-explainability-en-license). Adapted by FoodRadar. [Editorial standards](https://foodradar.org/editorial-policy)