Near-infrared hyperspectral imaging in tandem with partial least squares regression and genetic algorithm for non-destructive determination and visualization of Pseudomonas loads in chicken fillets.

Hyperspectral imaging was exploited for its potential in direct and fast determination of Pseudomonas loads in raw chicken breast fillets. A line-scan hyperspectral imaging system (900-1700 nm) was employed to obtain sample images, which were then further corrected, modified and processed. The prepared images were correlated with the true Pseudomonas counts of these samples using partial least squares (PLS) regression. To enhance model performance, different spectral extraction approaches, spectral preprocessing methods as well as wavelength selection schemes based on genetic algorithm were investigated. The results revealed that extraction of mean spectra is more efficient for representation of sample spectra than computation of median spectra. The best full wavelength model was attained based on spectral images preprocessed with standard normal variate, and the correlation coefficients (R) and root mean squared errors (RMSEs) for the model were above 0.81 and below 0.80 log10 CFU g(-1), respectively. In development of simplified models, wavelengths were selected by using a proposed two-step method based on genetic algorithm. The best model utilized only 14 bands in five segments and produced R and RMSEs of 0.91 and 0.55 log10 CFU g(-1), 0.87 and 0.65 log10 CFU g(-1) as well as 0.88 and 0.64 log10 CFU g(-1) for calibration, cross-validation and prediction, respectively. Moreover, the prediction maps offered a novel way for visualizing the gradient of Pseudomonas loads on meat surface. Hyperspectral imaging is demonstrated to be an effective tool for nondestructive measurement of Pseudomonas in raw chicken breast fillets.