Rapid identification and quantitative pit mud by near infrared Spectroscopy with chemometrics

Abstract

The properties of pit mud determine the quality of Chinese liquor. However, traditional methods for detection total carbon (TC), total nitrogen (TN) and total phosphorus (TN) in pit mud, such as Kjeldahl method, are time consuming and expensive. In this research, a time-saving and cost-effective method for identification and quantification of pit mud has been developed, which is based on near infrared spectroscopy (NIRS) combining with chemometrics. A total of 82 samples of different ages were collected from liquor-making factories and analyzed with NIRS (4000-10 000 cm-1). Four chemical pattern recognition methods including hierarchical cluster analysis (HCA), partial least squares-discriminant analysis (PLS-DA), artificial neural networks (ANN) and support vector machine (SVM) were used to build identification models of data set with aged and new pit mud. Four multivariate calibration methods, i.e., support vector regression (SVR), partial least squares regression (PLSR), ANN and extreme learning machine (ELM) were used to build quantitative model and compared for each data set, separately. Finally, in order to further improve the prediction accuracy, SG smoothing, derivatives, continuous wavelet transform (CWT), standard normal variate (SNV), and multiplicative scatter correction (MSC) were investigated. Results show that PLS-DA and SVM can achieve 100% classification accuracy for identification. PLSR and SVR were found to be optimal calibration methods for quantitative analysis of TC, TN and TP prediction in pit mud. Furthermore, the predictive ability can be improved by pretreating the spectra with optimal preprocessing method. The correlation coefficient (R) and root mean square error of prediction (RMSEP) of the independent validation reached to 0.9717 and 0.3491 for TC, 0.9862 and 0.0675 for TN, 0.9603 and 0.2162 for TP.