Abstract: A method based on near infrared (NIR) diffuse reflectance spectroscopy and chemometrics was developed to determine the content of lentinan.The spectral data of 60 samples were collected in the range of 12000-3800 cm^-1 and the content of polysaccharide were measured by UV-visible spectroscopic method.With the chemometric method,four outlier samples were found and eliminated,the NIR spectra were pretreated by various pretreatment methods and their combinations,and a wavelength method,stability competitive adaptive reweighted sampling (SCARS) was also used to select suitable informative wavelength points.After that,partial least square (PLS) calibration model was established with leave-one-out cross validation,and ten samples in the independent prediction set were predicted by the model.The results demonstrated that the predictive results of the PLS calibration model were optimum when the spectral data were pretreated by second derivative coupled with SCARS.At latent variable number of 10,the PLS model showed that correlation coefficient for calibration,RMSEC,correlation coefficient for prediction and RMSEP were 0.9906,0.0523 g/100 g,0.9781,0.0577 g/100 g,respectively.The model had a good predictive ability and can be used for the rapid determination of lentinan with NIR spectroscopy.

Key words: lentinan, NIR spectroscopy, partial least squares, SCARS