A Hybrid Method for Visual Near-Infrared Spectral Calibration for Discrimination of Panax Species

A novel method which was combination of uninformative variable elimination by partial least squares (UVE-PLS) and simulated annealing (SA) was proposed to extra relevant information among different varieties of panax. A total of 78 (26 for each variety) samples were characterized on the basis of visual and infrared spectroscopy (VIS-NIR), 63 (21 for each variety) samples were selected randomly for the calibration set, whereas, the remained 15 samples (5 for each variety) for validation set. UVE-PLS was executed to obtain the stability of each input variables. The best cutoff values of stability was determined as 34 using SA. Finally, 289 variables were mined, and inputted into partial least squares regression (PLSR) to build recognition model. The optimal model with the correlation coefficient of 0.9878, root mean square error of prediction (RMSEP) of 0.1297 was obtained. The performance was better than common PLS model which was established using the whole spectral wavelengths. The overall results indicated that the proposed method of UVE-PLS-SA was a powerful way to select diagnostic information for discrimination of different varieties of panax.