TVB-N prediction of tilapia with scales by information fusion of near infrared spectrum technology and sensory evaluation during chilled storage

The information fusion of near infrared spectrum technology and sensory evaluation was applied to predict the freshness of different parts for tilapia with scales during chilled storage.Spectral signatures ofbreast, middle and tail region in the range of 340 ~ 1063 nm were extracted.Smoothing Savitzky-Golay ( SG) , standard normal variate ( SNV) , polynomial derivative filters ( 1 st Der and 2 nd Der) were used for spectral pre-processing.Partial least square regression ( PLSR) was used to correlate the whole wavelengths spectra with total volatile basic nitrogen ( TVB-N) .Optimal wavelengths of different tilapia positions were selected by successive projections algorithm ( SPA) to develop new SPA-PLSR models, and the SPA-PLSR predictive performances of tails position ( root mean square error of prediction ( RMSEP) = 1.1295 mg/100 g, determination coefficient ( Rp²) = 0.8998) was better than that of breast and middle region, and also better than whole wavelengths model of tails region.Therefore, tail region was selected as spectrum sampling area.In order to evaluate the comprehensively fish freshness and improve the accuracy of model, spectral data and sensory evaluation were integrated for nondestructive measurement of freshness for tail region of tilapia based on PLSR, back-propagation artificial neural network ( BP-ANN) and least squares support vector machines ( LS-SVM) . Compared with single characteristic, information fusion of spectral data and sensory evaluation for LS-SVM had its superiority, which achieved accurate results with Rp² of 0.9255, RMSEP of 0.9701 mg/100 g.This result indicated that information fusion by integrating spectral data and sensory evaluation could significantly improve the TVB-N prediction performance, and it has tremendous potential in prediction of freshness in fish during chilled storage.