When monitoring the proximal composition of ground beef on a conveyor belt with a near infrared reflectance sensor, interference by signals from the belt itself proved to be a serious problem. A soft independent modelling of class analogies classification method proved useful in identifying spectra with belt components which, consequently, were removed to yield pure meat spectra. Partial lease squares regression on the basis of the corrected spectra yielded improved models for proximal analysis of the ground beef.
