基于部分线性回归的红外光谱多元校正方法

摘要

对于红外光谱数据而言,光谱-浓度关系常表现为一种复杂的混合线性关系.本文中提出一种部分线性回归算法,将复杂的光谱-浓度目标回归函数分解为线性和非线性决策函数之和.具体地,采用一序列的线性和非线性核函数来构建回归模型,分别用于逼近目标函数中的线性和非线性成分.本文中所提出的的方法与偏最小二乘回归算法和正则化最小二乘回归算法在3个实例数据集上进行实验对比.实验结果表明,本文中提出的算法具有更高的预测精度.%Spectra-concentrate relation is usually a very complex and mixed linear relation.In this paper,a partially linear regression (PLR) algorithm is proposed for multivariate calibration of spectroscopic data.In PLR,the target regression function is represented as the sum of several linear and nonlinear kernel decision functions, where each single kernel function with specific type and scale can approximate certain component of the target function.The proposed method is compared, in terms of RMSEP, with partial least squares regression (PLS) and regularized least-squares regression (RLS) method on three real spectroscopic data sets.Experimental results demonstrate that the proposed PLR method shows superiority over PLS and the single kernel RLS.