Abstract:To realize rapid and non-destructive detection of total volatile basic nitrogen(TVB-N) content of fresh beef with improved detection accuracy throughout the storage period at 4 ℃, a laboratory visible and near-infrared spectroscopy system was established to collect beef samples reflectance spectra between 400 and 1 700 nm. The beef samples were stored at 4 ℃ for 1 to 17 days. Multiplication scatter correction (MSC), first derivative (FD), Szvitzky-Golay(SG) smoothing method were used as pretreatment method for raw sample spectra. Combined with the effective wavelength variables extracted by uninformative variable elimination(UVE) and successive projections algorithm(SPA), the least square-support vector machine(LS-SVM) was proposed to predict beef TVB-N content. The results show that the SG is the best pretreatment method with reduced input variables by 99.5% for UVE and SPA, and the proposed LS-SVM has good performance with Rv of 0.925 and Sc of 4.615 mg?(100 g)-1,respectively.