Calibrating the Shelf-life of Chinese Flavored Dry Tofu by FTIR Spectroscopy and Chemometrics: Effects of Data Preprocessing and Nonlinear Transformation on Multivariate Calibration Accuracy

摘要

This paper demonstrates the application of FTIR spectroscopy coupled with chemometric methods to rapid analysis of the shelf-life of a traditional Chinese food, flavored dry bean curd (tofu). Transmittance FTIR spectra ranging from 4,000 to 400 cm−1 of 151 dry tofu samples were measured. Sample preparation involved finely grinding of samples and formation of thin KBr disks (under 10 MPa for 5 min). The shelf-life of dry tofu samples by spectroscopic analysis ranged from 29 to 161 days. Two different strategies were investigated to tackle the nonlinear problem: least squares support vector machines and partial least squares (PLS) with nonlinear transformation of spectral data by (a) B-spline function and (b) Gaussian kernel function. Different options of data preprocessing and calibration models were investigated to develop a nonlinear model with good generalization performance. Compared with linear models, both the two proposed strategies can improve the modeling and prediction accuracy with standard normal variate and second-order derivative spectra. B-spline-transformation PLS with second-order derivative spectra obtained a root mean squared error of prediction value of 9.6 days and was recommended as a simple-to-use nonlinear method for shelf-life prediction.