近红外光谱小波投影寻踪回归定量分析方法研究  

The Wavelet Projection Pursuit Regression Method of Near-infrared Spectroscopy Calibration

作　　者：周扬[1,2] 戴曙光[1] 吕进[2] 刘铁兵[2] 施秧[2] 陈才[2] 李博斌[3] 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093 [2]浙江科技学院,浙江杭州310023 [3]国家黄酒产品质量监督检验中心,浙江绍兴312071

出　　处：《计量学报》2013年第2期184-186,共3页Acta Metrologica Sinica

基　　金：国家自然科学基金（51075280）;国家质量监督检验检疫总局项目（2012（5））;浙江省自然科学基金（Y4110235）;浙江省重大科技专项和优先主题计划（2010C11060）;浙江省质量技术监督局项目（20110236）

摘　　要：为改进近红外光谱结构特征与定量回归模型的非线性拟合度和充分利用光谱中的非线性特征，提出了一种光谱小波投影寻踪定量分析方法。该方法对光谱进行小波分解后，用高斯混合模型噪声估计法降噪，对降噪后的小波系数向最优投影方向降维，用多项式岭函数拟合定量回归关系。建立黄酒近红外光谱快速预测酒精度小波投影寻踪回归模型，其相关系数R2和交叉检验标准差RMSECV分别为0．957和0．37838，该法比分析多元线性回归和偏最小二乘回归定量分析2种常规定量分析方法具有更优的预测效果，能更为有效地应用于近红外光谱快速定量分析检测。To improve the fitting degree between structural characteristics of near-infrared spectroscopy and non-linear quantitative regression model, a quantitative calibration method by the wavelet projection pursuit is proposed. The method reduces the noise of NIR wavelet coefficients by gaussian mixture model noise estimation method. After finding optimal low- dimensional projection of the spectrum wavelet coefficients, the calibration model was fitted by the polynomial ridge regression. The experiment sets up wine alcohol prediction model of near infrared spectroscopy by wavelet projection pursuit regression method, the R2 and RMSECV are 0. 957 and 0. 378 38, comparing with quantitative effect of multiple linear regression and partial least squares regression, spectral wavelet projection pursuit method has better prediction performance that can be more effectively applied quantitative analysis of near infrared spectroscopy.

关 键 词：计量学 定量分析 投影寻踪 近红外光谱 小波 

分 类 号：TB99[一般工业技术—计量学]