近红外光谱法快速测定香菇总糖含量

Application of Near Infrared Spectroscopy in the Rapid Detection of Total Sugar Content in Lentinula edodes

为弥补国标检测方法测定香菇总糖含量耗时长、步骤繁琐的缺陷,创建近红外(near infrared,NIR)光谱技术在测定香菇总糖含量方面应用,采用NIR分析技术与偏最小二乘算法(partial least square,PLS)建立香菇总糖的NIR分析模型,并对模型进行参数优化。实验共收集了106批样品,从中随机抽取13批作为验证集,用于验证该模型的可靠性,剩余的93批样品为校正集。在校正集中,通过杠杆值法和学生化残差法筛选出65批能够较理想地代表一般香菇特征的样品,用于确定NIR光谱检测范围、PLS主因子数等参数,基于NIR光谱数据的香菇总糖含量建立定量分析模型。校正集的建模结果表明,使用多元散射校正(multiplicative scatter correction,MSC)及二阶导数(second derivatives,SD)对原始光谱进行预处理,光谱范围为4 000~10 000 cm^(-1),PLS主因子数为10时,基于NIR的香菇总糖检测模型的建模效果最优,即均方根误差比值满足检验条件,校正集R^(2)(决定系数)最高为0.940 04,校正均方根误差为1.393,预测集均方根误差为1.557,相对分析误差最优为4.08。验证集对模型的检验结果显示,样品的预测值和实测值具有良好的线性关系,且二者没有显著差异(P=0.993)。由此表明,本实验建立的NIR分析模型可用于准确预测香菇样品的总糖含量。

In order to improve the defects of the method specified in the Chinese national standard to determine the total sugar content in edible fungi(Lentinula edodes),such as long detection time and tedious steps,near infrared spectroscopy(NIR)combined with partial least square(PLS)was adopted to establish a rapid method for determining the total sugar content of L.edodes and the predictive model parameters were optimized.In this study,a total of 106 samples were collected,13 of which were randomly assigned into the validation set for testing model reliability,while the remaining 93 were assigned into the calibration set.In the calibration set,according to the studentized residuals and leverage value,65 lots of samples with the typical characteristics of L.edodes were selected to optimize spectral range and the number of PLS factors for the establishment of a quantitative model for predicting the total sugar content in L.edodes.It was found that the optimal spectral preprocessing method was multiplicative scatter correction(MSC)combined with second derivatives(SD),and the predictive model with optimal results was constructed using the spectral range of 4000–10000 cm^(-1) and 10 PLS factors,with a root mean square error of calibration(RMSEC)of 1.393 and a root mean square error of prediction(RMSEP)of 1.557 as well as a correlation coefficient for calibration(R^(2))up to 0.94004 and a relative percent deviation(RPD)up to 4.08.Finally,the verification results showed the model-predicted value had a good linear relationship with the experimental value with no significant difference between them(P=0.993).In conclusion,the NIR prediction model could accurately predict the total sugar content of L.edodes.