摘要
以10 000～4 000cm-1波段的近红外光谱响应数据和常规生化方法检测的玉米蛋白质含量为样本数据,先对光谱响应数据进行小波去噪处理,并利用平滑技术对其降维,构建基于以光谱响应数据为输入、蛋白质含量为输出的偏最小二乘回归模型。仿真计算结果表明,利用偏最小二乘回归模型,可以较准确地预测玉米蛋白质含量,结合预测表达式回归系数和变量投影重要性指标VIP得到与蛋白质含量相关性较大的若干波段对应的光谱响应数据,模型在一定程度上揭示了蛋白质含量和光谱响应数据之间的数量关系。
Applying the partial least squares regression methodology, the protein contents of corn measured by using near infrared at 10 000--4 000 cm-1wave band and the conventional biochemical method were compared. The near infrared spectra were firstly de-noised and smoothed for dimensionality reduction, and then, the data was used as the input and protein content as the output for the model establishment. The simulation results showed that the protein content of corn could be accurately predicted by the regression model. The predictive expression regression coefficients and the variable importance in projection obtained on the spectra at a certain wavelengths highly correlated with the protein content. The model appeared to provide a significant quantitative relationship between the biochemical and near infrared determinations on the protein content of corn.
出处
《福建农业学报》
CAS
2014年第6期597-601,共5页
Fujian Journal of Agricultural Sciences
基金
国家自然科学基金项目(61101239)
关键词
偏最小二乘回归
近红外光谱
玉米蛋白质含量
小波去噪
partial least squares regression
near infrared spectra
protein content in corns
wavelet de-noising
作者简介
钟明洁（1993-），女，本科，研究方向：智能计算（E—mail：zhongmj0826@163．com）
通讯作者：王义康（1976-），男，博士，副教授，研究方向：复杂工业过程的建模与优化（E—mail：zjuwkang@163．com）
