基于蒙特卡洛变量组合集群分析法的小麦蛋白质近红外光谱变量选择

小麦是我国重要的粮食之一,提高对小麦蛋白质含量预测的精准性对保证小麦质量具有重要的意义。采用不同地区的小麦共93个样本近红外光谱与化学值作为建模数据。首先利用小波包(WTP)对光谱信号进行降噪处理,消除外界噪音信号对光谱的影响。其次利用蒙特卡洛变量组合集群分析法(MC-VCPA)进行变量选择,最后利用偏最小二乘法(PLS)建立小麦蛋白质预测模型,并对预测集样本进行预测。对比其他的建模方法,MC-VCPA所选择的10个特征变量代替了全光谱256个变量,预测均方根误差(RMSEP)值由0.4974降低到0.3295,提高了33%,优于其他建模方法。结果表明,基于MC-VCPA的近红外光谱分析方法对小麦蛋白质含量进行定量分析是可行的。

Variable Selection of Wheat Protein Near Infrared Spectra Based on Monte Carlo Variable Cluster Analysis

Wheat is one of the important grain in our country. It is important to improve the accuracy of wheat protein content prediction to ensure the quality of wheat,in this paper,the different parts of the wheat,a total of 93 samples of near infrared spectroscopy and chemical values as the modeling data. Firstly, the wavelet packet (WTP) is used to denoise the spectral signal to eliminate the influence of the external noise signal on the spectrum. Secondly, the Monte Carlo variable cluster analysis (MC-VCPA) method was used to select variables. Finally, the partial least squares (PLS) method was used to establish the wheat protein prediction model, and the forecast set samples were predicted. Compared with other modeling methods 10 variables selected by MC-VCPA instead of the full spectrum of 256 vari-ables, the root mean square error of prediction (RMSEP) value decreased from 0.4974 to 0.3295, increased by 33%, better than other modeling methods. The results show that it is feasible to do quantitative analysis of wheat protein content based on near infrared spectroscopy (MC-VCPA).