Abstract: In order to explore the feasibility of two-dimensional correlation synchronous spectra to select feature variables for meat freshness, visible/near infrared reflectance spectral information and total volatile basic nitrogen (TVB-N) content of 58 pork samples stored for 1–15 days were obtained. Then TVB-N content was employed as “external disturbance” and 10 representative spectra were selected for continuum removal. Seven spectral subregions were chosen according to the spectral difference and used for two-dimensional correlation analysis. By analyzing the synchronization spectra and the autocorrelation spectra, sensitive variables, which were closely related to TVB-N content, were obtained. Finally, using the selected variables, support vector machine (SVM) models for discrimination of pork freshness were established based on the original, standard normal variate preprocessed and normalized spectra, respectively. The results showed that 17 characteristic wavelengths, which accounted for only 1.61% of the total variables, were extracted by two-dimensional correlation spectral analysis, and that the overall accuracy rates of the SVM models were 94.83%, 98.28% and 98.28% respectively, indicating that the models performed well. Hence two-dimensional correlation analysis can be used to screen out the characteristic variables related to meat freshness. The research will be helpful for analyzing the change of spectral characteristics during meat spoilage and also provide new insights into variables selection in near infrared spectroscopy analysis.

Key words: visible/near-infrared spectroscopy, two-dimensional correlation spectrum, freshness, pork, discrimination model