摘要
基于Bayesian相似性评估方法结合偏最小二乘局部回归,对苹果近红外数据库进行数据挖掘。通过相似性计算方法搜索出与预测样品相近的近红外光谱,形成校正子集后采用局部回归方法获得待测样品的相关信息。该方法所建立局部模型的平均检验标准偏差(SEV)约为0.57,分析30个预测样品的预测标准偏差(SEP)约为0.61;基于马氏距离的传统方法建立的偏最小二乘局部模型的平均SEV为0.59,分析30个待测样品的预测SEP为0.64;而采用整个数据库建立的全局偏最小二乘模型的SEV约为0.65,分析30个预测样品SEP约为0.70。基于Bayesian相似性评估的局部回归方法在苹果糖度的近红外无损定量分析中获得较好的应用结果,在实际应用中该方法比全局回归方法具有更强的适用性,为近红外光谱分析提供了新的分析工具。
A novel local regression method combined with similarity evaluation for near infrared spectra was proposed.In this method,the similarity evaluation based on Bayesian statistics was utilized to compare the NIR spectra.The calibration subsets were then selected to construct the model by partial least square method according to the similarity.The sugar content of apple samples was predicted with the model.The mean values of the standard errors of validation（SEV） and prediction（SEP） for the partial least square local regression method based on Bayesian statistics（B-PLS） were 0.57 and 0.61,respectively.Those for the partial least square local regression method based on Mahalanobis distance（M-PLS） were 0.59 and 0.64,respectively,and those for the partial least square global regression method（G-PLS） were 0.65 and 0.70,respectively.The results showed that B-PLS could accurately predict the sugar content in apple and its performance was superior to that of G-PLS and a little superior to that of M-PLS.Thus the proposed method possesses higher accuracy compared with G-PLS and could be widely applied in the rapid and nondestructive analysis of internal qualities such as the sugar content of apple.Furthermore,the method could provide a new tool for near infrared analysis.
出处
《分析测试学报》
CAS
CSCD
北大核心
2010年第12期1173-1177,共5页
Journal of Instrumental Analysis
基金
国家863项目资助(2009AA04Z129)
关键词
近红外光谱
相似度
局部回归
糖度
苹果
near infrared spectra
similarity
local regression
sugar content
apple
