Abstract:

This study tested a new idea that beef freshness could be discriminated with multi-sensor information fusion from
machine vision and near infrared spectroscopy, meanwhile, successfully developed a system one can do this work on. In this
experiment, rump steak was used as experimental target. Based on feature variables from image information and spectral
information, back propagation (BP) neural network was adopted to establish a beef freshness classification model. Principal
component analysis (PCA) was implemented and the principal components (PCs) were extracted as the inputs, the freshness
divided by total volatile base nitrogen (TVB-N) contents as the outputs. Experimental results showed that the discrimination
rate equaled 98.31% under PCs = 6 for both image information and spectral information. The overall results showed that it
was feasible to discriminate beef freshness with two-sensor information fusion.

Key words: multi-sensor information fusion, beef freshness, nondestructive testing, machine vision, near infrared spectroscopy