基于VSPAI的近红外光谱小麦蛋白质分析方法研究  

Research on Near-Infrared Spectroscopy Wheat Protein Analysis Method Based on VSPAI

作　　者：张晓锋 张亦弛 宦克为[1] 金明杭 文鹏 ZHANG Xiaofeng;ZHANG Yichi;HUAN Kewei;JIN Minghang;WEN Peng(School of Physics,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学物理学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2024年第5期15-21,共7页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省科技发展计划项目(20240404046ZP)。

摘　　要：基于模型集群分析思想,利用变量选择比自适应迭代法(VSPAI)结合偏最小二乘法(PLS)建立了小麦蛋白质含量的近红外光谱预测模型。VSPAI利用蒙特卡罗采样法随机选取样本子集,运用PLS建立样本子集的回归模型,并计算各变量回归系数的平均值和标准差,求出每个变量的初始权重,并与加权自助采样法相结合选取最佳特征变量。结果表明,VSPAI-PLS模型与竞争性自适应重加权采样结合PLS、变量组合集群分析结合PLS、变量组合集群分析迭代保留信息变量法结合PLS、引导软阈值算法结合PLS模型相比,VSPAI-PLS模型的预测精度提高了39.9%、5.9%、25.4%、46.9%。综上,将VSPAI-PLS模型应用于小麦蛋白质的近红外光谱无损检测具备可行性。Based on the Model Population Analysis concept,a near-infrared spectroscopy prediction model for wheat protein content was established using the Variable Proportional Selection Adaptive Iteration(VSPAI)method combined with Partial Least Squares(PLS).VSPAI employs the Monte Carlo sampling method to randomly select sample subsets and uses PLS to build regression models for these subsets.The mean and standard deviation of the regression coefficients of each variable are calculated to obtain their initial weights,which are then combined with the Weighted Bootstrap Sampling method to select the optimal feature variables.The results show that the prediction accuracy of the VSPAI-PLS model improved by 39.9%,5.9%,25.4%,and 46.9%compared to the Competitive Adaptive Reweighted Sampling combined with PLS,Variable Combination Population Analysis combined with PLS,Variable Combination Population Analysis Iteratively Retained Information Variables combined with PLS,and Bootstrapping Soft Shrinkage combined with PLS models,respectively.In conclusion,the application of the VSPAI-PLS model for the non-destructive testing of wheat protein using nearinfrared spectroscopy is feasible.

关 键 词：小麦蛋白质 近红外光谱分析 模型集群分析 变量选择比自适应迭代 

分 类 号：O657.33[理学—分析化学]