Short-wavelength near-infrared (SW-near-IR) spectroscopy (700-1100 nm) is used for the determination of ethanol during the time course of a fermentation. Measurements are performed noninvasively by means of a photodiode array spectrometer equipped with a fiber-optic probe placed on the outside of the glass-wall fermentation vessel. Pure ethanol/water and ethanol/yeast/water mixtures are studied to establish the spectral features that characterize ethanol and to show that determination of ethanol is independent of the yeast concentration. Analysis of the second-derivative data is accomplished with multilinear regression (MLR). The standard error of prediction (SEP) of ethanol in ethanol/water solutions is approximately 0.2% over a range of 0-15%; the SEP of ethanol in ethanol/yeast/water solutions is 0.27% (w/w). Results from the mixture experiments are then applied to actual yeast fermentations of glucose to ethanol. By use of a gas chromatographic method for validation, a good correlation is found between the intensity of backscattered light at 905 nm and the actual ethanol. Additional experiments show that a calibration model created for one fermentation can be used to predict ethanol production during the time course of others with a prediction error of 0.4%.

Salt and moisture content of cured salmon roe (ikura) was determined using short-wavelength-near-infrared (SW-NIR) reflectance spectroscopy (600-1100 nm). SW-NIR can be used to measure chloride species. Calibrations for salt in bulk samples of high-quality roe (R(2) = 0.904, SEP = 0.421%, RPD = 3.21) and average-quality roe (R(2) = 0.711, SEP = 1.13%, RPD = 1.81), crushed eggs (R(2) = 0.857, SEP = 0.509%, RPD = 2.62), and individual eggs (R(2) = 0.731, SEP = 0.698%, RPD = 1.98) were developed using partial least squares (PLS) regression models. The heterogeneous distribution of lipid and moisture in the individual eggs limit the sensitivity of this method; however, this method provides a rapid nondestructive method for high-value food products where destructive testing is expensive or impractical and for process control applications.

Salt and moisture contents in cold-smoked salmon were determined using short-wavelength nearinfrared (SW-NIR) reflectance spectroscopy (600 to 1100 nm). Partial least square (PLS) regression models yielded the best results among 3 linear regression methods tested. Back-propagation neural networks (BPNN) exhibited a somewhat better capability to model salt and moisture concentrations (Salt: R 2 = 0.824, RMS = 0.55; Moisture: R 2 = 0.946, RMS = 2.44) than PLS (Salt: R 2 = 0.775, RMS = 0.63; Moisture: R 2 = 0.936, RMS = 2.65). Selection of samples from different axial locations on a fish did not affect the prediction error for salt or WPS but affected the prediction error for moisture.

No abstract