摘要：A method for the rapid and nondestructive determination of total nitrogen content in soy sauce was explored. Prediction models were established using near-infrared spectroscopy combined with each of the following techniques: partial least squares (PLS), interval PLS, synergy interval PLS (siPLS), and backward interval PLS. Results showed that each improved forecast model was better than the PLS model. The siPLS method exhibited the best performance. The full spectrum of light soy sauce was divided into 20 subintervals. The combination of four subintervals, namely, 6, 11, 13, and 18, showed the best effect, with a correlation coefficient of 0.9977 and an RMSECV of 0.0198. The full spectrum of dark soy sauce was divided into 20 subintervals. The combination of three subintervals, namely, 14, 17, and 19, showed the best effect, with a correlation coefficient of 0.9818 and an RMSECV of 0.0640. Therefore, the siPLS method can realise a rapid and accurate detection of total nitrogen content in soy sauce.
关键词：near-infrared spectroscopy; partial least squares
