Nondestructive Detection of Pork Tenderness Using Spatially Resolved Hyperspectral Imaging Technique Based on Multivariable Statistical Analysis
Sun Hong-wei,Peng Yan-kun,Wang Fan
DOI: https://doi.org/10.3964/j.issn.1000-0593(2019)11-3365-07
2019-01-01
Spectroscopy and spectral analysis
Abstract:Tenderness is one of the most important attributes of pork eating quality. The tenderness of pork depends on the complex physical and chemical characteristics of pork tissue. And a rapid, non-destructive detection method is urgently in need. This paper reports the feasibility of spatially resolved hyperspectral imaging technique for nondestructive detection of pork tenderness. First, the spatial resolved scattering images of 54 pork longissimus dorsi muscle were collected by hyperspectral system on line-scanning mode. The region of interest (ROI) was selected and the diffused spatial profile of incident light was extracted on the surface of the pork sample. The diffused spatial profile was fitted non-linearly by 4-parameter Lorentzian distribution function. The goodness of fit was R-2 > 0.992, and the residual analysis showed that the 4-parameter Lorentzian function could describe the spatial distribution of light intensity on meat surface. Four morphological parameters of spatial resolved spectrum at wavelength of 480 similar to 950 nm were extracted: asymptotic value a, peak value b, full width at half of the peak value c (FWHM) and slope at half of the peak value d. Partial least squares regression (PLSR) models were established to relate each parameter spectra and Warner Bratzler shear force (WBSF) values of pork samples respectively. The results showed that all parameters spectra contained pork tenderness information, in which the peak parameter b had the best prediction results, with determination coefficient of calibration set R-c(2) of 0.674, the root-mean-square error SEC of 8.396N, the determination coefficient of prediction set R-p(2) of 0.610, and the root-mean-square error SEP of 8.643N. In order to improve the accuracy and stability of the prediction model and realize the information fusion of multi-parameter spectra, PLSR analysis was firstly used to extract the latent variables in each parameter spectrum, which have high relative variance contribution to pork tenderness. Then, the latent variable scores were combined as the characteristic variables of the parameter spectra, and multiple statistical regression analysis was performed to relate the characteristic variables and the WBSF values of pork samples. In order to avoid data redundancy, PLSR algorithm was secondly used to reduce and transform the characteristic variables of the parameter spectra. Using the cross validation method, the first two-dimensional factor scores were selected to establish the calibration model. The variance interpretation rate of the pork WBSF value from the first factor was 92.28%. Compared with the PLSR model built by the single-parameter spectrum, the prediction results of the multi-parameter spectra model have been greatly improved, with R-c(2) of 0.923 and R-p(2) of 0.800, SEC of 4.083N and SEP of 5.655N respectively. The results show that all regression coefficients are very significant (p<0.01). In this study, the multi-parameter information fusion method was adopted to provide an idea for the application of spatial resolution spectroscopy in the nondestructive testing of pork tenderness. This method decomposed the spatial resolved spectra into 4 morphological parameters effectively, and achieved the information extraction and fusion of different parameter spectra, providing technical support for the development of non-destructive rapid detection equipment for pork tenderness based on spatial resolved spectroscopy technology.