Abstract:

In this research spectral information for commercially available chicken breast samples was extracted in the
medium wavelength region of near-infrared (NIR) and used for establishing a quantitative analysis model of cholesterol
concentration using partial least squares (PLS) regression, and the prediction accuracy of the model was analyzed. The
modeling was investigated with respect to outlier elimination and optimized combined use of various spectral preprocessing
methods. The optimum quantitative analysis model was achieved with two cycles of outlier elimination and the combined
use of SG first derivative, SG smoothing and De-trending for spectral preprocessing. The model parameters standard error of
calibration (SEC), correlation coefficient of calibration set (Rc), standard error of prediction (SEP), correlation coefficient of validation
set (Rp), ratio of performance to standard deviate (RPD), relative standard error of prediction (RSEP), the number of principal factors (PF)
and SEP/SEC were 4.534 8, 0.919 7, 7.437 5, 0.812 0, 2.844 7, 9.44%, 5, and 1.640 1, respectively. The experimental results showed
that the model had better prediction accuracy in test set (P > 0.05), especially in the range of 60–100 mg/100 g, and could be applied to
analyze the cholesterol concentration of commercially available fresh chicken and chicken products.

Key words: near-infrared spectroscopy, medium wavelength region, fresh chicken, cholesterol concentration, quantitative analysisscopy