Quantitative estimation of cold storage time of meat using hyperspectra data

The objective of this paper is to develop a method for estimating the cold storage time of pork using hyperspectra image cubes data accessed by a self-developed Filed Imaging Spectrometer System (FISS). Data were preprocessed by Minimum Noise Fraction (MNF) transform and first derivative to eliminate high-frequency random noise and baseline offset and to improve the multicollinearity. Wilks´ lambda stepwise method was performed to select proper wavelengths and partial least square regression (PLSR) was executed to build the model and get the regression equation. Fisher Linear Discriminant Analysis (LDA) was applied to discriminate the cold storage time of pork. The total of 600 samples was chosen, half of them as training samples and half as validation ones. The results showed that overall distinguishing rate reached 99% with seven selected bands of fresh or frozen-thawed pork samples. For the cold storage time, classification accuracy was higher than 83% with eleven selected bands. It indicates that FISS might be used as a screening method to identify the quality of frozen meat.