摘要
为解决近红外光谱快速检测乳品成分及含量时光谱数据的预处理问题,提出一种基于直方图分层映射技术的近红外光谱主成分得分重置(SR)预处理方法。以葡萄糖氯化钠水溶液三组分样品中的葡萄糖含量、鲜牛奶样品中的乳糖含量为定量检测目标,进行散射光谱主成分得分累计贡献率的分层分段规定化映射预处理,利用偏最小二乘(PLS)回归分析建模手段,对相应近红外光谱中的糖含量信息进行测试及分析。结果表明,经过SR预处理后,牛奶中乳糖含量PLS模型的校正集样品交互验证预测偏差降低23.9%,实际预测偏差降低27.8%;验证集实际预测偏差降低16.7%。该SR光谱预处理方法兼顾光谱、参考值及组分相关性等多尺度信息,以实现光谱信息增强去噪,能避免有用信息误删,防止不充分拟合及过拟合。
In order to solve the pre-processing problem of spectral data in near infrared rapid detection analysis of content of milk components,apre-processing algorithm for score resetting（SR）of principal components（PC）in the near infrared spectrum on the basis of histogram layering mapping technology is proposed.With glucose content in the three components samples consisting of glucose,NaCl and water,and lactose content in the fresh milk samples,as the detecting objects,cumulative contribution rates of the near infrared scattering spectral PC scores are pre-processed by means of mapping by layer and by piece.Furthermore,partial least squares（PLS）regression analysis method is used for modeling,thereby test and analysis of sugar content information in corresponding near infrared spectra are completed.The results show that after SR pretreatment,the predicted deviation of the calibration curve of the milk lactose content PLS model is reduced by 23.9%,the actual prediction deviation is reduced by 27.8%,and the actual prediction deviation of the verification set is reduced by 16.7%.This SR spectral preprocessing method takes into account multi-scale information such as spectra,reference content value and component correlation to realize spectral information denoising enhancement.Therefore,false deletion of useful information can be avoided,and inadequate fitting and overfitting can be prevented.
出处
《激光与光电子学进展》
CSCD
北大核心
2017年第9期387-395,共9页
Laser & Optoelectronics Progress
基金
国家自然科学基金(11574065)
关键词
光谱学
光谱分析
光谱预处理
主成分得分重置
直方图分层映射
spectroscopy
spectrum analysis
spectral pre-processing
principal component score resetting
histogram layering mapping
