波长选择SCARS和偏最小二乘法建立香菇中多糖含量预测的近红外光谱模型

Establishment of Near Infrared Spectral Model for the Prediction of the Content of Lentinan Using Wavelength Selection SCARS and Partial Least Square Analysis

为实现香菇多糖含量的快速测定,利用近红外光谱漫反射技术采集了60个香菇粉末样本在12000~3800 cm-1范围内的光谱数据,利用紫外可见光谱法测定了香菇粉末样品的多糖含量。采用多种化学计量学方法,剔除掉四个异常样本后,考察了不同的光谱预处理方法以及波长选择对模型的影响,用留一交互检验法建立了偏最小二乘(PLS)模型,并用所建立的校正模型对独立预测集样本进行了预测。结果表明,当采用二阶导数及变量稳定性的竞争自适应加权抽样法(SCARS)选择的波长对光谱进行处理时,所建立的模型预测效果最佳,在隐变量数为10时,模型相关系数为0.9906,校正均方根误差(RMSEC)为0.0523 g/100 g,预测相关系数Rp=0.9781,预测均方根误差(RMSEP)=0.0577 g/100 g,该模型具有较好的预测能力,可用于香菇多糖含量的近红外光谱快速检测。

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.