Predicting intramuscular fat content variations in boiled pork muscles by hyperspectral imaging using a novel spectral pre-processing technique

For better predicting intramuscular fat contents in pork muscles using hyperspectral imaging, a novel correlation optimised warping (COW) technique was employed with the first derivative on the full spectra and the feature wavelengths selected by successive projections algorithm. Images of 104 pork longissimus dorsi samples cooked in boiling water for eight different periods were taken using a Vis-NIR (400-1000 nm) imaging system. Duplex method was used to divide the images into training and predicting sets. Reference measured intramuscular fat contents of each sample were correlated with the spectra extracted from ROI within the corresponding samples. Support vector regression models were developed and results proved positive effects by COW combined with first derivative transforms as spectral pre-processing techniques. The simplified model developed based on eight important wavelengths (403, 435, 438, 556, 586, 596, 739 and 951 nm) predicted accurately the intramuscular fat contents with R-P(2) of 0.9635 and RMSEP of 0.885 g/kg. Some other algorithms were used and listed as control algorithms to enhance data analysis, including Savitzky Golay (SG)-smoothing, standard normal variate (SNV), multiplicative scatter correction (MSC) and partial least squares regression (PLSR).