南丰蜜桔可溶性固形物近红外检测的多元线性回归变量筛选  

Selection of Multiple Linear Regressions Variables for Determination Soluble Solids in Nanfeng Mandarin by Near-Infrared Spectroscopy

作　　者：孙旭东[1] 郝勇[1] 蔡丽君[1] 刘燕德[1] 

机构地区：[1]华东交通大学机电工程学院光机电技术及应用研究所,南昌330013

出　　处：《生物物理学报》2011年第8期727-734,共8页Acta Biophysica Sinica

基　　金：国家科技支撑计划(2008BAD96B04);江西省对外科技合作计划(2009BHB15200);江西省主要学科学术和技术带头人培养对象计划(2009DD00700)~~

摘　　要：文章采用反向区间偏最小二乘法结合连续投影算法,筛选南丰蜜桔近红外检测的多元线性回归变量。对南丰蜜桔近红外光谱进行多元散射校正后,利用反向间隔偏最小二乘法,从500～1750 nm中初选出7个光谱区间,用于多元线性回归变量筛选。利用通过遗传算法和连续投影算法筛选出的变量建立了多元线性回归模型。经比较发现,利用反向区间偏最小二乘法结合连续投影算法筛选出的变量建立的多元线性回归模型,预测结果最优,模型预测相关系数为0.937,模型预测均方根误差为0.613 oBrix。结果表明,反向区间偏最小二乘法结合连续投影算法,可以有效地筛选近红外光谱的多元线性回归变量,提高南丰蜜桔可溶性固形物模型的预测精度。Near infrared（NIR） spectroscopy variables for multiple linear regressions（MLR） were chosen to predict soluble solids in Nanfeng mandarin based on backward interval partial least squares（BiPLS） and successive projections algorithm（SPA）.The spectra of Nanfeng mandarin were pretreated by multiplicative scatter correction（MSC）.Seven spectral intervals were chosen in the wavelength range of 500～1750 nm.The intervals were applied to select variables for MLR.MLR models were developed by the variables selected by genetic algorithm（GA） and SPA.By comparison,the model was the best calibrated with BiPLS-SPA variables.The correlation coefficient of prediction（Rp） was 0.937,and the root mean square error of prediction was 0.613 oBrix.The results suggested that BiPLS-SPA was an effective method to select NIR variables,and improve the precision of MLR models for predicting soluble solids of Nanfeng mandarin.

关 键 词：光谱学 近红外 连续投影算法 反向区间偏最小二乘法 变量筛选 多元线性回归 

分 类 号：O657.3[理学—分析化学]