Abstract
In-line monitoring of the proximal composition of ground beef on a conveyer belt has been tested using a near infrared (NIR) reflectance instrument with a diode array detector. Sixty batches of coarsely ground beef were processed under industry conditions and monitored continuously. After removing signals originating from the belt itself, the remaining data were used to make partial least squares models for all samples at two different grinding sizes. The correlation coefficients were in the range 0.93–0.96, while the full cross-validated errors for fat and water were between 1.6 and 2.4%. The corresponding errors for protein were 0.5–0.8%. A forward variable selection method based on jack-knifing yielded similar results. The predictions were generally best for the smallest grinding size. The loading plots showed that there was a high interdependency between calibrations for the different components. Before implementation of the calibrations, they must be re-tested under industry conditions on new independent batches.
Get full access to this article
View all access options for this article.