基于偏最小二乘留一交叉验证法的近红外光谱建模样品选择方法的研究

目的提出一种新的挑选定标集的方法-偏最小二乘留一交叉验证法。方法以玉米为例,通过对玉米中水分含量的实际建模与外部验证,根据主成分数、相关系数、预测均方根差以及相对分析误差（ratio of performance to standard deviate,RPD）等因素,综合比较4种定标集挑选方法的优缺点。结果偏最小二乘留一交叉验证法结合样品和光谱性质,在保持原始样品覆盖范围的基础上,挑选出的定标集所建立的模型具有较低的模型复杂程度、较高的验证相关系数以及较高的RPD值。结论该方法既克服了随机挑选法存在的样品代表性不足的风险,同时也避免了含量梯度法和计算机识别法只考虑样品或者光谱的单一性质的不足...

Full description

Saved in:
Bibliographic Details
Published in食品安全质量检测学报 Vol. 8; no. 1; pp. 182 - 186
Main Author 白鹏利 王钧 尹焕才 殷建 田晶晶 陈名利 高静
Format Journal Article
LanguageChinese
Published 中国科学院苏州生物医学工程技术研究所,苏州,215163 2017
Subjects
偏最小二乘留一交叉验证法
定标集
样品挑选
近红外光谱
partial least squares-leave one out-cross validation
偏最小二乘留一交叉验证法
近红外光谱
样品挑选
sample selection
calibration set
near infrared spectroscopy
定标集
Online AccessGet full text

Cover

More Information
Summary:目的提出一种新的挑选定标集的方法-偏最小二乘留一交叉验证法。方法以玉米为例,通过对玉米中水分含量的实际建模与外部验证,根据主成分数、相关系数、预测均方根差以及相对分析误差（ratio of performance to standard deviate,RPD）等因素,综合比较4种定标集挑选方法的优缺点。结果偏最小二乘留一交叉验证法结合样品和光谱性质,在保持原始样品覆盖范围的基础上,挑选出的定标集所建立的模型具有较低的模型复杂程度、较高的验证相关系数以及较高的RPD值。结论该方法既克服了随机挑选法存在的样品代表性不足的风险,同时也避免了含量梯度法和计算机识别法只考虑样品或者光谱的单一性质的不足,同时该方法具有操作简单、易于推广等优点,为食品安全检测提供了一种新的筛选样品的方法。
Bibliography:near infrared spectroscopy; partial least squares-leave one out-cross validation; sample selection; calibration set
BAI Peng-Li,WANG Jun,YIN Huan-Cai,YIN Jian,TIAN Jing-Jing,CHEN Ming-Li,GAO Jing（Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Science, Suzhou 215163, China）
Objective To establish a new method of selecting calibration set-partial least squares-leave one out-cross validation（PLS-LOO-CV）. Methods According to the values of principal component number, correlation coefficient（R）, root mean square error of prediction（RMSEP）, and ratio of performance to standard deviate（RPD）, the 4 selection methods of calibration sets were compared by modeling and validation of water content in rice. Results By PLS-LOO-CV combined with samples and spectral properties, the selected sets of the model had lower model complexity, higher correlation coefficient and higher RPD value on the basis of maintaining the original sample coverage. Conclusion The established method can avoid the risk o
ISSN:2095-0381