Abstract:The storage time of salmon is one of the critical indexes to evaluate salmon quality, while the traditional methods are costly, destructive and time-consuming. This study highlights the effectiveness of Raman spectroscopy combined with support vector regression (SVR) and variable selection method in the prediction of salmon storage time. The spectra of salmon fillets during storage at -18℃, 4℃ and 20℃ were collected by the portable Raman spectrometer. To decrease the model complexity and improve the model prediction accuracy, three spectral variable selection algorithms were investigated for model optimizing, including competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA) and hybrid method (CARS-SPA). Comparing the full spectral models with the models established based on the variables selected by above variable selection methods, the results showed that CARS-SPA coupled with SVR (CARS-SPA-SVR) produced the optimal prediction performance. The determination coefficient and root mean square error of prediction set (R2p and RMSEP) at -18℃, 4℃ and 20℃ can reach 0.9986 and 0.1606, 0.9988 and 0.1485, 0.9995 and 0.0371, respectively. Therefore, this study have proved that Raman spectroscopy coupled with CARS-SPA-SVR model is an effective strategy for rapidly, nondestructively and accurately predicting the storage time of salmon fillets stored at different temperature.

Raman spectroscopy combined with support vector regression and variable selection method for accurately predicting salmon fillets storage time