ABSTRACT
The predicted model of the sensory quality of refrigerated tilapia skin at 8°C was constructed using near-infrared spectroscopy (NIRS) based on the principle of sample packaging integrity. The characteristic spectral intervals were chosen from preprocessed spectral data using synergy interval partial least squares (siPLS), principal component analysis, and Jordan–Elman back propagation-artificial neural network (JENN) was used to build a prediction model for tilapia skin quality parameter. The results showed that the multiple scatter correction (MSC) + 2nd derivative was the best-preprocessed method, and the four characteristic spectral intervals were 680–740, 742–800, 980–1040, and 1150–1210 nm. Further, the cumulative contribution rate of the first three principal components was 99.15%. Additionally, the root mean square error of cross-validation set (RMSECV) of transfer function (tanh) of the model was 0.386, the determinant coefficient for prediction () was 0.973, and the RMSECV and
were 0.393 and 0.971 for unknown samples, respectively. The results showed that NIRS combined with JENN could allow rapid and accurate evaluation of tilapia skin quality in the range of 73.00–97.00 scores.
Disclosure statement
No potential conflict of interest was reported by the authors.