独立成分分析支持向量机回归模型及其在近红外光谱分析中的应用

Independent Component Analysis-Support Vector Regression and Its Application in Near Infrared Spectral Analysis

首先采用独立成分分析(ICA)提取近红外光谱数据矩阵的独立成分和相应的混合矩阵,然后用支持向量机回归(SVR)对混合矩阵和实测浓度矩阵进行建模,建立了独立成分分析-支持向量机回归(ICA SVR)的近红外分析建模方法.结果表明,ICA SVR模型的预测结果明显优于SVR和偏最小二乘法(PLS)方法,方法用于肉样品中水分、脂肪和蛋白质的同时测定,获得了满意的结果.

A new model building method of near-infrared （NIR） spectra based on independent component analysis （ICA） and support vector regression （SVR） was proposed. In this method, independent components matrix and the corresponding mixing matrix can be extracted from the original NIR spectra by ICA, then SVR was used to build a model between mixing ma- trix and the concentration matrix of chemical components. It was observed that the correlation between different independent components and chemical concentrations were obviously different. After a selection of independent components in the modeling process, the prediction results can be improved effectively. The effect of this method was validated by its application in the quantitative prediction of moisture, fat and protein of meat samples.