Fusion of spectra and texture features of hyperspectral imaging for quantification and visualization of characteristic amino acid contents in beef

Characteristic amino acids is an important indicator for evaluating the nutritional value and flavor parameter of beef. To accurately quantify and visualize of arginine and alanine content in beef, the feasibility of visible nearinfrared hyperspectral multivariate calibration analysis and data fusion was explored. Three methods were used to select the optimal feature wavelength and fusion multi-level texture information, and to develop the prediction performance of linear, non-linear and neural network models in different signals. Compared to all models, the linear model demonstrated superior performance in terms of characteristic spectral and fused spectral. Among, the more effective prediction performances emerged from CARS-ASM-ENT-PLSR model with RP2 = 0.9211, RMSEP = 0.1252 mg/100 g and RPDp = 3.49 for alanine. The UVE-ASM-ENT-HOM-COR-PLSR model for arginine prediction achieved a good performance of RP2 = 0.8596, RMSEP = 0.8596 mg/100 g and RPDp = 2.62. Finally, visualization plots of arginine and alanine content distribution were generated. This study shows that the data fusion method provides a new approach for rapid evaluation of CAA content.