Visualizing quantitatively the freshness of intact fresh pork using acousto-optical tunable filter-based visible/near-infrared spectral imagery

Abstract

Although pork freshness is one of the top concerns to consumers, no systems are currently available to the pork industry that could quantitatively predict its spatial distribution in a rapid and nondestructive way. The main objective of this study was to investigate the feasibility of acousto-optical tunable filter (AOTF) based spectral imagery in the visible/near-infrared region for the non-destructive prediction and visualization of the spoilage-indicating chemicals over the surface of intact fresh pork. We developed an AOTF-based spectral imaging system (wavelength range: 550–1000nm) to visualize pork freshness by mapping the predicted total volatile basic nitrogen (TVB-N) content over the surface. Reflectance hyperspectral images of pork loins in packages (n=43) were acquired from day 3 to day 13 post-mortem, and the corresponding TVB-N references were recorded using conventional chemical procedures. The eligible muscle region of interest (EMROI) on a sample surface was auto-segmented, from which the signature spectrum was extracted. After standard normal variate (SNV) filtering, the signature spectra together with their chemical references were fed into a partial least squares regression (PLSR) to create a prediction model on a consecutive spectral range (575–940nm). An analysis of the regression coefficients identified 9 important predictive wavelengths (575, 600, 615, 705, 765, 825, 885, 915, and 935nm). The prediction model was subsequently refined to use the feature wavelengths only. A leave-one-out (LOO) cross-validation showed that the prediction of the TVB-N contents using the refined model was good and had a root mean square error (RMSECV) of 1.94mg/100g and a coefficient of determination (Rcv2) of 0.89. Finally, the freshness distribution over an entire pork surface was visualized by mapping the pixel-wise TVB-N predictions in pseudo-colors based on the refined model. The spatial prediction was also verified in terms of mean and range. The mean values coincided well with their chemical references (with a R2 of 0.81 and a RMSE of 2.58mg/100g), and the range is within reasonable limits (with 95% pixels within 0-50.0mg/100g). The results indicated that the AOTF-based spectral imagery system could be a promising method to predict pork freshness in an in situ test with unprecedented details of the spatial distribution of freshness.Industrial relevance: An AOTF-based VIS/NIR spectral imagery system has the potential for acceptance sampling in meat production plants or for hygienic supervision in the marketplace to predict the freshness of intact chill-stored pork.