Hyperspectral imaging can make pork eating quality a usable specification

A study published on 14 October proposes visible/near-infrared hyperspectral imaging as a way to classify the eating quality of pork shoulder butt. Its relevance lies in the fact that it does not seek to predict a single analytical value; instead, it groups samples according to an overall eating-quality profile. The researchers combined sensory-panel and electronic-tongue results, then defined eating-quality clusters through statistical clustering.

The dataset comprised 189 samples. The images were segmented into muscle, fat and whole-meat regions. The muscle region delivered the strongest classification performance, both for a two-class distinction and for a three-class model. According to the study abstract, the team also evaluated feature-selection methods and data augmentation in combination with different machine-learning algorithms.

The practical significance is a shift in the question being asked: it is not only about detecting colour, fat or water, but about approaching an expectation of eating pleasure. That said, the camera does not become a sensory judge by itself. Reference clusters depend on the sample set, panel training, product and cooking method. Before being used for commercial sorting, a plant would need to validate the model against its own animals, cuts, fabrication methods, packaging and intended culinary use.

For processors and retailers, the approach offers an interesting perspective for shoulder products, pulled pork, roasts and marinated lines: batches with comparable eating profiles could be directed more precisely to the right application. The condition is that prediction remains tied to a clear specification and regular sensory verification. Imaging can accelerate a decision; the quality promise still has to be proven on the plate.