Fourier transform near-infrared spectroscopy coupled with variable selection methods for fast determination of salmon fillets storage time

Abstract

To address the fast and nondestructive determination of salmon fillets storage time associated with its freshness, Fourier transform near-infrared (FT-NIR) spectroscopy coupled with advanced variable selection methods was attempted in this work. Fresh salmon fillets were divided into three groups and stored at -18℃, 4℃ and 20℃, respectively. The spectra under each temperature were collected by a FT-NIR spectrometer with the range of 4000–11000 cm−1. Advanced variable selection methods, including random frog (RF), successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), CARS combined with SPA (CARS-SPA) and variable combination population analysis combined with iteratively retaining informative variables (VCPA-IRIV), were utilized to extract key feature variables and shrink variable space. The prediction models were constructed by using partial least squares regression (PLSR) based on full spectral variables and selected feature variables. Variable selection method VCPA-IRIV coupled with PLSR (VCPA-IRIV-PLSR) showed the best prediction performance in each temperature, with determination coefficients of prediction set (R2P) of 0.9988, 0.9976 and 0.9998, and root mean square error of prediction set (RMSEP) of 0.145, 0.209 and 0.024 for -18℃, 4℃ and 20℃, respectively. The overall results showed the great feasibility of FT-NIR combined with variable selection methods in the rapidly, nondestructively and accurately predicting freshness of salmon fillets stored at different temperatures.