Simultaneous multi-component analysis of pork meat during bacterial spoiling process by FT-NIR evaluated with a non-linear algorithm

The feasibility was studied of simultaneously determining, by Fourier transform near infrared (FT-NIR) spectroscopy combined with a non-linear algorithm, the chemical components of pork meat which had been stored at a temperature of 4 °C. The spectra of 120 pork meat samples stored at 4 °C were measured during an 11 days spoiling process. Synergy interval partial least square (SI-PLS) was performed to select characteristic spectral variables of different components in pork meat based on NIR spectral data preprocessed by standard normal variate (SNV). Meanwhile, the characteristic spectral variables of different components of pork meat were determined by principal component analysis (PCA), and the top principal components (PCs) were extracted as the input of a back-propagation artificial neural network (BP-ANN) model, respectively. The BP-ANN model was optimized by cross-validation, and the optimum BP-ANN model was determined according to the lowest root mean square error cross-validation (RMSECV). Experimental results showed that the BP-ANN models for predicting different components content in pork meat are superior to SI-PLS regression models, in which the coefficient of determination (Rp2) and root mean square error of prediction (RMSEP) for predicting the contents of total fat, total sugar, protein and TVB-N in pork meat were 0.874, 0.899, 0.897, 0.856 (Rp2) and 0.426 g per 100 g, 0.016 g per 100 g, 3.305 g per 100g, 5.141 mg per 100 g (RMSEP), respectively. The overall results demonstrate that NIR spectroscopy combined with BP-ANN, as a nondestructive analytical tool, could be utilized in order to determine the quality and freshness of pork meat and related products.