Abstract:

Acetaldehyde and acetal, aroma compounds important to the balance and harmony of liquor have a significant
impact on its flavor and quality. In order to accurately and quickly measure acetaldehyde and acetal in base liquor, calibration
models were established based on Fourier-transform near infrared (FT-NIR) spectroscopy with artificial neural network by
cross-validation after the spectra were analyzed. The results showed that the acetaldehyde and acetal in liquor base exhibited
a strong characteristic absorption in the near infrared (NIR) region. The best spectral pretreatment methods were found to be
first derivative + vector normalization (SNV) and first derivative + multivariate scatter correction (MSC). According to the
selected spectral ranges of 7 501.7–7 102.0, 6 101.7–5 446.0 cm-1 (for acetaldehyde) and 6 101.7–5 449.8 cm-1 (for acetal),
the internal cross validation method was adopted to establish a model. Through optimization of the model, the coefficients
of determination (R2) between chemically measured values and NIR predicted values in calibration set for acetaldehyde
and acetal were 0.963 4 and 0.942 9, respectively with internal cross-validation root mean square deviation of 0.50 and
0.83 mg/100 mL, respectively. The prediction model for acetaldehyde and acetal was validated. It was demonstrated
that the R2 values for acetaldehyde and acetal in validation set were 0.975 0 and 0.918 6, respectively with prediction
standard deviation of 0.80 and 1.0 mg/100 mL, respectively. The method has been applied to quick determination of the
concentrations of acetaldehyde and acetal with satisfactory results in the distilled spirit industry.

Key words: Fourier transform near infrared spectroscopy, base liquor, acetaldehyde, acetal