Non-destructive analysis of salt and moisture in food products by short-wavelength near-infrared (SW-NIR) spectroscopy.

摘要

Short-wavelength Near Infrared (SW-NIR) (600–1100 nm) reflectance spectroscopy was used to non-destructively analyze salt and moisture content in cured and/or smoked fish products. Linear regression methods: multiple linear regression (MLR), principal components regression (PCR) and partial least square regression (PLS), plus a non-linear back-propagation neural networks (BPNN) were used to correlate SW-NIR spectra with reference values. Linear and non-linear models were developed to measure salt and moisture in cured and/or smoked salmon products. The accuracy of both the linear and non-linear models are sufficient to make adoption of the SW-NIR method practical in the aquatic food processing industry.; Temperature fluctuation during spectral measurement is one of the major sources of prediction error in NIR models. Pure aqueous NaCl solutions (0–10% w/v) at temperatures of 4.0–42.9°C were used to study the temperature effects. PLS and BPNN methods were used for SW-NIR calibration to determine sample temperature and salt concentration, respectively. A global model was built for temperature determination. Three different model systems: a global model, a multiple temperature model system, and a room temperature model with correction factors were built for salt concentration determination. Global models provided better results than multiple temperature models or room temperature models, but global models needed a large sample size for model development. Room temperature models with appropriate correction factors may yield results close to that of global models.