基于变量组合集群分析法的小麦蛋白质近红外光谱变量选择方法研究

为了解决小麦蛋白质的近红外光谱信息复杂、共线性严重及全光谱建模的预测能力不足等问题,采用一种新的变量选择方法——变量组合集群分析法(VCPA)对小麦蛋白质的近红外光谱进行特征波长选取.首先利用二进制矩阵采样策略(BMS)和指数衰减函数(EDF)删除无信息变量,优选小麦中蛋白质近红外特征波长,然后结合偏最小二乘法(PLS)建立预测模型.与其他变量选择方法相比,VCPA所选用的波长点最少,模型的预测能力最强,VCPA算法所采用的BMS变量采样策略弥补了蒙特卡洛采样方法的不足.研究结果表明,VCPA算法可以有效选择小麦蛋白质近红外光谱特征波长,提高预测模型的可靠性和适用性.

Research on Variable Selection of Protein in Wheat Near Infrared Spectroscopy Based on Latent Projective Graph

In order to solve the near infrared spectra of wheat protein complex information, collinearity serious and full spectrum of modeling prediction ability is insufficient,a new method of variable selection is adopted what variable com-bination population analysis (VCPA) on the near infrared spectrum characteristics of wheat protein wavelength selec-tion. Based on the features of the binary matrix sampling (BMS) and uninformative variable elimination strategy of index decreasing function (EDF), VCPA explored optimally the efficient wavelength from the NIR spectroscopy the wheat to develop models for prediction the protein of the wheat. The results showed that the performance of VCPA model was superior to the performances from others selection variables method with the least variable. Good prediction performance was obtained for protein of wheat. the BMS variable sampling strategy made up for the deficiency of the Monte Carlo sampling method. The study demonstrated that VCPA could effectively select the characteristic wave-lengths of NIR spectral to improve the model robustness and applicability.