Non-destructive detection of TVB-N content in fresh pork based on hyperspectral imaging technology

Non-destructive detection of total volatile basic-nitrogen (TVB-N) content in fresh pork based on hyperspectral imaging (HSI) technology was evaluated. In order to build a firm hyperspectral calibration model of TVB-N content in fresh pork, discussion about the impacts of different data processing methods(sample set partitioning, characteristic wavelength selection, and so on)on the predictive results was done. The results showed that, the best partial least square regression (PLSR) calibration model established with the whole spectral range was that using concentration gradient (CG) to partition samples, and using normalize combined with mean center(MC) to preprocess the spectral, resulting in R
_cv
²
of 0.9508 and R
_p
²
of 0.9357, RMSECV of 1.3042 and RMSEP of 1.5130. After 49 optimal wavelengths were selected by using competitive adaptive reweighted sampling (CARS) algorithm, 93.5% data redundancy was got rid of and the precision of PLSR model which was established using the selected 49 wavelengths was improved. It increased R
_c
v
²
to 0.9640, R
_p
²
to 0.9531, and decreased RMSECV to 1.1151, RMSEP to 1.2885. All the results showed that hyperspectral imaging technology is a powerful technique to detect the TVB-N content in fresh pork non-destructively.